DENG 2022 Archivage

THÈSE
Pour obtenir le grade de

DOCTEUR DE L’UNIVERSITE GRENOBLE ALPES
Spécialité : 2MGE: Matériaux, Mécanique, Génie civil,
Electrochimie
Arrêté ministériel : 25 mai 2016
Présentée par
Na DENG
Thèse dirigée par François NICOT, et
Codirigée par Antoine WAUTIER, Yannick THIERY et
Zhen-Yu YIN
préparée au sein du Laboratoire INRAE ETNA

dans l'École Doctorale I-MEP2 - Ingénierie - Matériaux,
Mécanique, Environnement, Energétique, Procédés,
Production
Micromécanique de l'état critique et son

émergence dans la modélisation multi-
échelle de la rupture des sols
Micromechanics of Critical State and its
emergence in multiscale modeling of soil
failure
Thèse soutenue publiquement le 17 janvier 2022,
devant le jury composé de :
Monsieur Stefan LUDING
PROFESSEUR, University of Twente (Netherlands), Rapporteur
Madame Xia LI
PROFESSEUR, Southeast University (China), Rapporteuse
Monsieur Ali DAOUADJI
PROFESSEUR, INSA Lyon (France), Président
Monsieur Gioacchino VIGGIANI
PROFESSEUR, 3SR (France), Examinateur
Monsieur Félix DARVE
PROFESSEUR EMERITE, 3SR (France), Examinateur
Monsieur Pierre-Yves HICHER
PROFESSEUR EMERITE, Ecole Centrale de Nantes (France), Examinateur
Monsieur François NICOT
DIRECTEUR DE RECHERCHE, INRAE, Université Savoie Mont-Blanc (France),
Directeur de thèse
Monsieur Antoine WAUTIER
Chercheur, INRAE (France), Invité
Monsieur Yannick THIERY
Chercheur, BRGM (France), Invité
Monsieur Zhen-Yu YIN
Professeur associé, The Hong Kong Polytechnic University (China), Invité
“We cannot solve our problems with the same thinking we
used when we created them.”
Albert Einstein
Acknowledgements
First and foremost, I would like to express my sincere gratitude to Dr. François Nicot
and Dr. Antoine Wautier for their clear guidance and continuous encouragement during
the three years. I really appreciate the wonderful research environment they created in
our group through organizing in-group and out-group meetings regularly. I have learnt so
much from them, from technical methods to critical thinking skills. I have been inspired
to keep an open mind and stay curious. I know the experience working with them will
benefit me for life.
I would like to extend my sincere gratitude to my co-supervisors Dr. Yannick Thiery

and Dr. Zhen-Yu Yin for their support and confidence in my work. They always responded
quickly when I asked for help.
I am very grateful to Dr. Pierre-Yves Hicher and Dr. Félix Darve for their valuable
suggestions and insightful discussions during the three years. They monitored the progress
of the thesis as committee members and defense jury members. I must also thank other
jury members Dr. Stefan Luding, Dr. Xia Li, Dr. Gioacchino Viggiani and Dr. Ali
Daouadji for their careful reading of my manuscript and for their helpful comments and
rich discussions during the defense. It is a great honor for me to defend my thesis in front
of them.
Thanks should also go to Dr. Antoinette Tordesillas for broadening my understanding

of granular materials from physical and statistical perspectives, to Dr. Luc Sibille for the
interesting discussions and to my master supervisor Dr.Yangping Yao for the fundamental
training in soil mechanics and his support to apply for the opportunity to study abroad.
I also appreciate the technical help from Dr. Jiaying Liu, Dr. Hao Xiong and Dr. Zhuang
Jin during the three years. Special thanks to Ms. Pearl-Angelika Lee, Shimu, for valuable
remarks on English writing of journal manuscripts.
I am thankful to French Geological Survey (BRGM) and China Scholarship Council

(CSC). The financial supports under BRGM Contract No. RP19DRP023 and under CSC
Grant No. 201801810030 are acknowledged.
I would like to thank my group-mates and friends Adriane Clerc, Tao Wang and Marie
Miot. It has been a great pleasure working and spending time with them. Big thanks to
5
6
my friends Lingling Xuan, Xinming Xing, Dou Quan, Yanning Hu, Bo Ren and Qiangqiang
Sun for enjoyable and unforgettable time with them after work during the three years.
In the end, my sincere gratitude to my family for their unwavering love and support all
the time and deep thanks to Jiangfeng with whom I can share all my emotions, thoughts
and dreams.
Abstract
Multiscale modeling of granular materials has received an increasing attention in recent

years based on notable progress in instrumentation technologies and computing power. It
will probably play a prominent role in the future of constitutive modeling as an alternative
method to phenomenological models in large-scale engineering problems, especially those
concerned by heterogeneity and strain localization, e.g., landslides. In multiscale modeling,
the main challenge is to simulate continuum-scale behavior using information obtained at
finer scales. As for the continuum-scale behavior in granular materials, critical state is one
of the most important features that manifests in drained triaxial tests by the capacity of
a geomaterial to accommodate shear deformation under constant stresses and without any
volume change.
A fresh perspective is presented in this thesis for the standard concept of critical state
(CS) in granular materials by suggesting that CS can be defined using a single proportional
strain test. A comparison between proportional strain tests and biaxial tests simulated with
discrete element methods (DEM) shows that the CS line characterized by the stress state
(effective pressure and deviatoric stress), void ratio and fabric indexes can act as an at-
tractor. The mechanical responses and fabric metrics evolve along dilatant proportional
strain loading paths according to similar values after the strain level has become large
enough to wipe out the initial material memory in the homogeneous domains considered in
this analysis, i.e., the shear band domain in dense samples subjected to strain localization
and the whole sample domain in loose samples. This suggests that the contact network
of a granular material subjected to proportional strain loading evolves while preserving its
ability to withstand shearing without volume change. Therefore, the CS concept can be
found in a wide class of loading paths which shows that CS acts as a general attractor
irrespective of the loading path considered.
Once critical state is reached, a stationary regime takes place in a biaxial test. The
macroscopic state remains constant while the microstructure is constantly rearranging. In
this thesis, we examine Critical State in granular materials from a dynamical systems the-
ory (DST) perspective. The aim is to enrich the classical view of the critical state regime
in granular materials, and more broadly, improve fundamental understanding of the un-
derlying mesoscale mechanisms responsible for macroscopic stationary states in complex
systems. The lifespan and life expectancy of specific cluster conformations comprising par-
ticles in force chains and grain loops, are tracked and quantified. Results suggest that these
conformational clusters reorganize at similar rates in the critical state regime, depending
on strain magnitudes and confining pressures. We quantified this rate of reorganization
i
ii
and found that the material memory rapidly fades, with an entirely new generation of force
chains and grain loops replacing the old ones within a few percent strain.
Based on these extensive investigations on the critical state, mesostructural transfor-

mations are found which bring vanishing and generating mesostructures along the loading
path as a central ingredient that should be included in multiscale constitutive modeling. A
deactivation/reactivation procedure acting on the local mesoscale is proposed in a specific
micromechanical-based constitutive model (the H-model). Preliminary results show that
this is a promising direction to improve the model to predict relatively large strains and
to make stationary regimes emerge naturally without drawing from any empirical law at
the macroscopic scale.
Keywords: Granular materials, Critical state, Proportional loading path, Dynamical

system, Microstructure, Multiscale modeling, DEM, Micromechanics
Résumé
La modélisation multi-échelles des matériaux granulaires a fait l'objet d'une attention crois-
sante ces dernières années, grâce aux progrès notables de des techniques d'instrumentation
et à l'augmentation des puissances de calcul. Il est probable qu'elle jouera un rôle essentiel
à l'avenir en tant que méthode alternative aux modèles phénoménologiques dans les prob-
lèmes d'ingénierie à grandes échelles, en particulier ceux qui concernent l'hétérogénéité et
la localisation des déformations, par exemple, dans la modélisation des glissements de ter-
rain. Dans la modélisation multi-échelle, le principal défi est de simuler le comportement à
l'échelle du continuum en utilisant au maximum les informations obtenues aux échelles plus
fines dans le système, plutôt que de recourir à des modèles constitutifs empiriques. En ce qui
concerne le comportement à l'échelle du continuum dans les matériaux granulaires, l'état
critique est l'une des caractéristiques les plus importants, que l'on définit généralement
dans des essais triaxiaux drainés comme la capacité pour un géomatériau à encaisser de
la déformation de cisaillement à état de contrainte constant et sans changement de volume.
Une approche originale est présentée dans cette thèse concernant concept classique
d'état critique dans les matériaux granulaires en suggérant que l'EC peut être défini par
l'utilisation d'un seul essai en déformation proportionnelle. Une comparaison entre des
essais en déformation proportionnelle et des essais biaxiaux simulés par la méthode aux
éléments discrets (DEM) montre que la ligne d'état critique caractérisée par l'état de con-
trainte (pression effective et contrainte déviatorique), l'indice des vides et le déviateur du
tenseur de texture peut agit comme un attracteur. Toutes les réponses mécaniques et les
indicateurs de microstructure évoluent lors des chemins de chargement dilatés à déforma-
tion proportionnelle selon des valeurs similaires après que le niveau de déformation soit
suffisamment important pour effacer la mémoire initiale du matériau dans les domaines
homogènes considérés dans cette analyse, c'est-à-dire la zone de la bande de cisaillement
dans les échantillons denses soumis au phénomène de localisation de la déformation et
l'échantillon tout entier dans les échantillons lâches. Cela suggère que la microstructure
d'un matériau granulaire soumis à un chargement de déformation proportionnelle évolue
avec une capacité constante à résister au cisaillement sans changement de volume. Par
conséquent, le concept d'état critique peut être retrouvé dans une large classe de chemins
de chargement, puisqu'il agit comme un attracteur, quel que soit le chemin de chargement
considéré.
Une fois l'état critique atteint, un régime stationnaire peut se mettre en place dans
un essai biaxial. L'état macroscopique reste inchangé alors que la microstructure se réar-
range en permanence. Dans ce travail de thèse, nous examinons ces régimes stationnaires
iii
iv
dans les matériaux granulaires du point de vue de la théorie des systèmes dynamiques
(DST). L'objectif est d'enrichir la vision classique de l'état critique dans les matériaux
granulaires et, plus largement, d'améliorer la compréhension fondamentale des mécanismes
mésoscopiques sous-jacents responsables des états stationnaires macroscopiques dans les
systèmes complexes. La durée de vie et l'espérance de vie de conformations spécifiques de
clusters, comprenant quelques particules comme les chaînes de force et les cycles de grains,
sont suivies et quantifiées. Les résultats suggèrent que ces structures se réorganisent à des
vitesses similaires à l'état critique, proportionnellement aux amplitudes de déformation et
aux pressions de confinement considérées. Nous avons quantifié ces taux de réorganisation
et constaté que la mémoire du matériau s'estompe rapidement, une génération entièrement
de nouvelles chaînes de force et cycles de grains remplaçant l'ancienne en quelques pour-
centages de déformation.
Sur la base de ces études micromécaniques détaillées sur l'état critique, on constate que
les transformations mésostructurales correspondant à la disparition et la génération de mé-
sostructures au cours du chargement est un ingrédient central qui devrait être inclus dans
les modèles multi-échelles du comportement constitutif des matériaux granulaires. Une
procédure de désactivation/réactivation de mésostructures est proposée dans un modèle
micromécanique particulier : le modèle H. Les résultats préliminaires montrent que cette
procédure constitue une direction prometteuse pour améliorer le modèle et faire émerger
naturellement un régime stationnaire sans s'appuyer sur une loi empirique à l'échelle macro-
scopique.
Mots clés : Matériaux granulaires, État critique, Chemin de chargement proportion-

nel, Systèmes dynamiques, Microstructure, Modélisation multi-échelle, DEM, Micromé-
canique.
Contents
1 General introduction 1
1.1 Motivation and objectives . . . . . . . . . . . . . . . . . . . . . . . . . . . 1
1.2 Structure of presentation . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3
2 A brief literature review 5
2.1 Shearing behavior of granular materials . . . . . . . . . . . . . . . . . . . . 5
2.1.1 Dilatancy and contractancy . . . . . . . . . . . . . . . . . . . . . . 6
2.1.2 Softening and hardening . . . . . . . . . . . . . . . . . . . . . . . . 8
2.1.3 Critical State . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 10
2.2 Discrete element view of granular materials . . . . . . . . . . . . . . . . . . 11
2.2.1 Discrete Element Method . . . . . . . . . . . . . . . . . . . . . . . 11
2.3 Micromechanical analysis in granular materials . . . . . . . . . . . . . . . . 15
2.3.1 Particle/contact . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 16
2.3.2 Force chains . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 16
2.3.3 Grain loops . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 18
2.4 Multiscale modeling of granular materials . . . . . . . . . . . . . . . . . . . 20
3 On the attraction power of critical state 23
3.1 DEM simulation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 24
3.2 Mechanical response . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 27
3.2.1 Mechanical response along dilatant proportional strain loading paths 27
3.2.2 Mechanical response along biaxial loading paths . . . . . . . . . . . 28
3.3 p-q-e space analysis . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 30
v
vi Contents
3.3.1 p-q plane . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 30
3.3.2 p-e plane . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 38
3.4 Fabric-related critical state locus . . . . . . . . . . . . . . . . . . . . . . . 39
3.4.1 Fabric tensor analysis . . . . . . . . . . . . . . . . . . . . . . . . . . 39
3.4.2 Grain loop evolution . . . . . . . . . . . . . . . . . . . . . . . . . . 41
3.5 Mixed proportional strain and biaxial loading paths . . . . . . . . . . . . . 46
3.6 Conclusion and outlook . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 52
4 Dynamical view of the critical state 55
4.1 Numerical setup . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 58
4.2 The so-called critical state . . . . . . . . . . . . . . . . . . . . . . . . . . . 58
4.2.1 Stress-strain analysis . . . . . . . . . . . . . . . . . . . . . . . . . . 58
4.2.2 Force chains and deviatoric stress . . . . . . . . . . . . . . . . . . . 59
4.2.3 Grain loops and volumetric strain . . . . . . . . . . . . . . . . . . . 61
4.3 Hidden dynamics at critical state . . . . . . . . . . . . . . . . . . . . . . . 63
4.3.1 Generating and vanishing process of chained particles . . . . . . . . 63
4.3.2 Generating and vanishing process of grain loops . . . . . . . . . . . 69
4.4 Microstructure reorganization dynamics under different confining pressures 69
4.5 Microstructure dynamics under evolving conditions: memory effects in gran-

ular materials . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 81
4.5.1 Fading process of the initial memory . . . . . . . . . . . . . . . . . 82
4.5.2 Memory fading process along proportional strain test . . . . . . . . 87
4.6 Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 93
5 Critical state and the H-model 95
5.1 Review and analysis of the H-model . . . . . . . . . . . . . . . . . . . . . . 97

Contents vii
5.1.1 The H-model in brief . . . . . . . . . . . . . . . . . . . . . . . . . . 97
5.1.2 Biaxial test at the material point scale . . . . . . . . . . . . . . . . 102
5.1.3 Mesoscale inspection of the H-model during biaxial loading . . . . . 105
5.2 Emergence of critical state in the H-model . . . . . . . . . . . . . . . . . . 114
5.2.1 Method . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 114
5.2.2 Preliminary results . . . . . . . . . . . . . . . . . . . . . . . . . . . 115
5.3 Conclusions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 121
6 General conclusion and perspectives 123
6.1 Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 123
6.1.1 Dilatant proportional strain loading and relationships with critical

state . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 123
6.1.2 Dynamical view of critical state . . . . . . . . . . . . . . . . . . . . 124
6.1.3 Enriched H-model with emerging critical state . . . . . . . . . . . . 125
6.2 Perspectives . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 125
6.2.1 Micromechanical investigations . . . . . . . . . . . . . . . . . . . . 125
6.2.2 H-model . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 128
6.2.3 Multiscale modeling of landslides . . . . . . . . . . . . . . . . . . . 129
A Complementary results on critical state dynamics 133
Bibliographie 146
List of Figures
2.1 Typical deviatoric stress and volumetric strain responses in a dense (left)
and a loose (right) granular assembly under a biaxial loading path. . . . . 6
2.2 images of a system of photoelastic disks for an isotropically compressed state

(left) and a sheared state (right) (Majmudar and Behringer 2005) . . . . . 12
2.3 Avatar specimen under isotropic stress (left) and at the end of triaxial load-
ing (right) (Kawamoto et al. 2018). . . . . . . . . . . . . . . . . . . . . . . 12
2.4 Elasto-frictional contact law: components in the contact (left), behavior in

the tangential direction (right). . . . . . . . . . . . . . . . . . . . . . . . . 15
2.5 Force chain definition (Peters et al. 2005). . . . . . . . . . . . . . . . . . . 17
2.6 Grain loops categorized by edge length in 2D granular assembly. . . . . . . 19
3.1 Quasi-2D DEM specimens for (a) proportional strain tests where loading
parameters are ε1 and ε2 and (b) biaxial tests where loading parameters are
ε2 and σ2 . The direction 2 is always the vertical loading control direction at
the REV scale. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 27
3.2 Evolution of the deviatoric stress q along dilatant proportional strain tests
with different magnitudes of dilatancy characterized by λ = −1.2, −1.3, −1.4.
Point I, point II and point III refer to onset of a well marked shear band
at ε2 = 0.010, slightly blurred shear band at ε2 = 0.109, and the pressure
p close to zero at ε2 = 0.142, respectively. The corresponding kinematic
patterns of these three points are shown in Fig.3.3 . . . . . . . . . . . . . . 30
3.3 Incremental deviatoric strain maps (dεd ) estimated for axial strain incre-
ments of 0.11%. Three axial strains are considered along the proportional
strain path λ = −1.2: (I) onset of a well marked shear band at ε2 = 0.010,
(II) slightly blurred shear band at ε2 = 0.109, (III) just before liquefaction
(p is close to 0) at ε2 = 0.142. The corresponding points in q − ε2 plane are
shown in Fig.3.2 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 31
3.4 Deviatoric stress and volumetric strain of biaxial tests in dense (left panel)
and loose (right panel) samples. Three biaxial loading paths with σ0 =
40 kPa, 60 kPa and 100 kPa are considered. Note that soil mechanics
convention is adopted with positive compression. . . . . . . . . . . . . . . . 32
ix
x List of Figures
3.5 Principal stresses in the whole sample (σ1 and σ2 ) and within the shear
band (σ1∗ and σ2∗ ), and relative orientation of principal stress (θσ∗1 − θσ1 ).
The 100kPa-dense sample under biaxial loading path is adopted. . . . . . . 33
3.6 Stress paths of biaxial tests and proportional strain tests in p − q plane.
The start and end points are marked by triangles and squares, respectively.
The critical stress ratio curve and the maximum stress ratio curve are drawn
according to the critical states and the maximum value from biaxial tests,
respectively. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 35
3.7 Stress paths in p −q plane along six proportional strain tests and one biaxial
test. Dilatant (λ = −1.2), undrained (λ = −1) and contracting (λ = −0.5)
proportional strain paths are conducted with the samples labelled 100kPa-
dense and 100kPa-loose. A biaxial loading path is performed in the sample
100kPa-dense. The square domain is zoomed up on the right hand side. . . 36
3.8 Four categories of stress path along proportional strain tests according to
DEM simulation results shown in Fig.3.7. . . . . . . . . . . . . . . . . . . . 36
3.9 p−e∗ evolution in proportional strain (a) and biaxial (b) tests. The start and
end points are marked by triangles and squares, respectively. A fit based
on a power function is shown for proportional strain tests and repeated
to compare with the results from biaxial tests, as all dilatant proportional
strain tests converge towards a master curve. The equation of the fitted
curve is e∗ = 0.2571 − 0.0227( 100
p 0.6274
) . . . . . . . . . . . . . . . . . . . . . 37
3.10 p − αc∗ evolution for proportional strain (a) and biaxial (b) tests. The start
and end points are marked by triangles and squares, respectively. A linear
fit is shown for the dilatant proportional strain tests and repeated on the
biaxial tests, as all the proportional strain tests converge towards a master
curve. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 40
3.11 p − r3∗ for proportional strain (a) and biaxial (b) test. The start and end
points are marked by triangles and squares, respectively. An exponential
curve. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 42
3.12 p − r6p∗
for proportional strain (a) and biaxial (b) tests. The start and end
curve. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 43
List of Figures xi
biaxial tests, as the all proportional strain tests converge towards a master
curve. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 44
curve. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 45
3.15 p − q evolution along combining proportional strain and biaxial loading

paths. The proportional strain loading with λ = −1.2 are conducted at first
until a certain axial strain level labeled as A, B, C, D, E and F, after that
a biaxial loading is performed by keeping the corresponding lateral stress
unchanged. The transition states and the average critical states are marked
by dots and squares, respectively. The corresponding stress-strain responses
are shown in Fig.3.16 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 47
3.16 Stress-strain responses along combined proportional strain and biaxial load-
ing paths. The switch states and ultimate states are marked by dots and
squares, respectively. The corresponding stress paths are shown in Fig.3.15 48
3.17 The relation between imposed dilatancy from proportional strain test and
unconstrained dilatancy if switched to a biaxial condition according to re-
sults in Fig.3.15 and Fig.3.16. Note that soil mechanics convention is adopted
with positive compression. . . . . . . . . . . . . . . . . . . . . . . . . . . . 49
3.18 The relation between imposed dilatancy/contractancy in proportional strain

tests and unconstrained dilatancy/contractancy expected in biaxial tests in
the plane of axial strain vs. volumetric strain (left) and the possible stress
paths in the plane of deviatoric stress vs. mean stress (right). The major
solid lines represent the imposed dilatancy/contractancy, namely the pro-
portional strain loading paths; the fins denote the incremental unconstrained
dilatancy/contractancy if the loading is switched to biaxial conditions. . . 49
4.1 Evolution of deviatoric stress q and volumetric strain εv with respect to the
axial strain ε2 along biaxial tests. Note that soil mechanics convention is
adopted with positive compression and positive contraction. . . . . . . . . 60
xii List of Figures
4.2 Incremental deviatoric strain (IDS) maps at the critical states, estimated for
axial strain increments of 0.1%. Two samples are considered: (a) 100kPa-
dense sample (ε2 = 6.1%) and (b) 100kPa-loose sample (ε2 = 8.02%). . . . 61
4.3 Proportions of chained particles.Two samples are considered: (a) 100kPa-

dense sample and (b) 100kPa-loose sample. Note that rcha ∗
represents the
value when the spatial domain of shear band are considered, and rcha when
the whole sample is considered. The deviatoric stresses are recalled. . . . . 62
4.4 Volumetric strain of different types of loops and the global assembly during
the biaxial tests in 100kPa-dense sample (a) and 100kPa-loose sample (b). 64
4.5 Lifespan and life expectancy PDFs (probability density functions) of chained
particles in two samples: (a) 100kPa-dense sample where only the spatial
domain within the shear band is considered and (b) 100kPa-loose sample
with the whole domain considered. For each sample, two reference confor-
mations belonging to the stationary regime are adopted, labeled as M and
N. The deviatoric stress curve (dot-dashed line) and the volumetric strain
(dashed line) curves are recalled. . . . . . . . . . . . . . . . . . . . . . . . 66
4.6 The averaged PDFs of chained particles with respect to the incremental
strain evolution (∆ε2 ), over six microstructure conformations at different
strain levels belonging to the critical regime (ε2 = 0.0610, 0.0708, 0.0806,
0.0904, 0.1002 and 0.1100 for the 100kPa-dense sample and ε2 = 0.0605,
0.0708, 0.0802, 0.0906, 0.1005 and 0.1106 for the 100kPa-loose sample). . . 67
4.7 The strain magnitudes within the shear band domain (100kPa-dense*),
whole dense sample (100kPa-dense) and the whole loose sample (100kPa-
loose). The strain magnitude ratio between 100kPa-dense* and 100kPa-
dense is around 2.8. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 68
4.8 PDFs for chained particles from 100kPa-loose sample and from 100kPa-dense
sample stretched by the strain magnitude ratio 2.8. . . . . . . . . . . . . . 68
4.9 Lifespan and life expectancy PDFs (probability density functions) of grain
loops in two samples with respect tho the axial strain ε2 : (a) 100kPa-dense
sample where only the spatial domain within the shear band is considered
and (b) 100kPa-loose sample with the whole domain considered. For each
sample, two reference conformations belonging to the stationary regime are
adopted. The deviatoric stress curve (dot-dashed line) and the volumetric
strain (dashed line) curves are recalled in the subfigure of Loop 4. . . . . . 70
List of Figures xiii
4.10 The averaged PDFs of grain loops with respect to the incremental strain
evolution (∆ε2 ), over six microstructure conformations at different strain
levels belonging to the critical regime (ε2 = 0.0610, 0.0708, 0.0806, 0.0904,
0.1002 and 0.1100 for the 100kPa-dense sample and ε2 = 0.0605, 0.0708,
0.0802, 0.0906, 0.1005 and 0.1106 for the 100kPa-loose sample). . . . . . . 71
4.11 PDFs for grain loops from 100kPa-loose sample and from 100kPa-dense sam-
ple stretched by the strain magnitude ratio 2.8. . . . . . . . . . . . . . . . 72
4.12 Lifespan and life expectancy PDFs for chained particles under different con-
fining pressures with samples 40kPa-loose, 100kPa-loose and 400kPa-loose
considered. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 73
4.13 Lifespan and life expectancy PDFs for grain loops with respect to the in-
cremental strain evolution (∆ε2 ) under different confining pressures with
samples 40kPa-loose, 100kPa-loose and 400kPa-loose considered. . . . . . . 74
4.14 Double-exponential fits given for the chained particle PDFs with respect to
the absolute value of incremental evolution |∆ε2 | under different confining
pressures. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 76
4.15 Double-exponential fits given for the loop PDFs with respect to the absolute
value of incremental evolution |∆ε2 | under different confining pressures. . . 77
4.16 Two characteristic strains (εc1 and εc2 ), the characteristic probability (Pc )
and the ratio εεc1
c1
with respect to the mean stress at the critical state. The
critical mean stress under different confining pressures (40 kPa, 100 kPa and
400 kPa) are 59 kPa, 143 kPa and 559 kPa, respectively. . . . . . . . . . . 78
4.17 Color maps of the lifespan of grains loops for the reference axial strain
ε2 = 8.02%. Two thresholds are considered corresponding to the two char-
acteristic strains: (a) 0.4% and (b) 3%. . . . . . . . . . . . . . . . . . . . . 80
4.18 Color maps of the life expectancy of grains loops for the reference axial
strain ε2 = 8.02%. Two thresholds are considered corresponding to the two
characteristic strains: (a) 0.4% and (b) 3%. . . . . . . . . . . . . . . . . . . 80
4.19 Life expectancy PDFs (probability density functions) of chained particles at

the initial state of the whole sample in the dense (a) and loose (b) cases. The
deviatoric stress curve (dot-dashed line) and the volumetric strain (dashed
line) curves are recalled. . . . . . . . . . . . . . . . . . . . . . . . . . . . . 83
xiv List of Figures
4.20 Life expectancy PDFs (probability density functions) of grain loops at the
initial state of the whole sample in the dense (a) and loose (b) cases. The
deviatoric stress curve (dot-dashed line) and the volumetric strain (dashed
line) curves are recalled in the subfigure of Loop 4. . . . . . . . . . . . . . 85
4.21 Color map of life expectancy of grain loops at the initial state under the
biaxial test in the dense (a,b) and loose (c,d) samples. Two thresholds are
used in the color scale: 10 (a,c) and 120 (b,c) times the reference incremental
axial strain δε1 = 0.1% which corresponds respectively to 1 % and 10 % of
axial strain from the initial state in the biaxial loading . . . . . . . . . . . 86
4.22 Life expectancy PDFs (probability density functions) of grain loops under
the dilatant proportional strain loading considering the whole sample. . . . 88
4.23 Lifespan and life expectancy PDFs (probability density functions) of grain
loops at the convergent state under the dilatant proportional strain loading
considering a domain within the shear band. . . . . . . . . . . . . . . . . . 89
4.24 The six states chosen for color maps along the proportional strain tests . . 90
4.25 The absolute value of PDFs for Lifespan and life expectancy of grain loops
under the dilatant proportional strain loading within the domain of the shear
band. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 90
4.26 The absolute value of PDFs for Lifespan and life expectancy of grain loops
under the dilatant proportional strain loading with the whole sample con-
sidered. A fit is given to the PDFs using the double exponential in Equation
4.3. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 91
4.27 Color map of life expectancy of grain loops at different conformations: (a)
the initial state ε2 = 0, (b) the peak of deviatoric stress ε2 = 0.0185, (c-f)
belonging to the convergent stage under the dilatant proportional strain test
in the dense sample ε2 = 0.0610, 0.0810, 0.0910, 0.1094. . . . . . . . . . . . 92
5.1 The directional distributions and the coordinates at the REV scale (e1 , e2 )
and the mesoscale (n, t). . . . . . . . . . . . . . . . . . . . . . . . . . . . . 96
5.2 The symmetrical description of geometry (a) and forces (b) at the elementary
hexagonal pattern of adjoining particles (H-cell). There are two types of
contact c1 and cc , the corresponding contact branches are d1 and d2 . . . . . 96
5.3 General homogenization scheme relating both incremental stress and strain
tensors . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 97
List of Figures xv
5.4 Three options of the meso-volume: hexagonal domain (blue), rectangular

domain going through the grain centers (green), bounding box domain (red) 101
5.5 Two modes of pathological H-cell due to contact loss. . . . . . . . . . . . . 103
5.6 Two modes of over-compressed H-cell with the corresponding opening angle. 103
5.7 The stress ratio and volumetric strain curves based on the parameters from
Veylon 2017. Three confining pressures are considered. . . . . . . . . . . . 104
5.8 The deviatoric stress and volumetric strain along the biaxial loading with
p0 = 100 kPa until a relative large axial strain. . . . . . . . . . . . . . . . . 105
5.9 Evolution of the ratio between the length d1 and its initial value d0 : (a)
directional analysis at different axial strains ε11 = 0.0, 0.005, 0.032, 0.128 and
0.347 labelled as O, A, B, C and D, respectively, (b) macro strain evolution
along specific directions. . . . . . . . . . . . . . . . . . . . . . . . . . . . . 106
5.10 Evolution of the ratio between the length d2 and its initial value d0 : (a)
directional analysis at different axial strains ε11 = 0.0, 0.005, 0.032, 0.128 and
0.347 labelled as O, A, B, C and D, respectively, (b) macro strain evolution
along specific directions. . . . . . . . . . . . . . . . . . . . . . . . . . . . . 107
5.11 Porosity ϕ1meso of the hexagon with respect to the opening angle α; sketch
of the two possible deforming processes. . . . . . . . . . . . . . . . . . . . . 108
5.12 Evolution of the opening angle α: (a) directional analysis at different axial
strains ε11 = 0.0, 0.005, 0.032, 0.128 and 0.347 labelled as O, A, B, C and D,
respectively, (b) macro strain evolution along specific directions. . . . . . . 109
5.13 Evolution of Vmeso σ11meso

: (a) directional analysis at different axial strains
ε11 = 0.0, 0.005, 0.032, 0.128 and 0.347 labelled as O, A, B, C and D, respec-
tively, (b) macro strain evolution along specific directions. . . . . . . . . . 110
5.14 Evolution of Vmeso σ22meso

ε11 = 0.0, 0.005, 0.032, 0.128 and 0.347 labelled as O, A, B, C and D, respec-
tively, (b) macro strain evolution along specific directions. . . . . . . . . . 111
5.15 Schematic model to reset α with the porosity being kept. The reset 1 is
triggered when the opening angle smaller than 60◦ tends to decrease further,
and the reset 2 when the opening angle larger than 60◦ tends to increase
further. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 115
5.16 Flowchart of the computation of biaxial loading incorporating the deactiva-

tion/reactivation scheme. . . . . . . . . . . . . . . . . . . . . . . . . . . . . 116
xvi List of Figures
5.17 Deviatoric stress and volumetric strain with respect to the axial strain using
the updated H-model. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 117
5.18 Polar diagrams of the ratio between the length d1 and its initial value d0 : at
different axial strains ε11 = 0.000, 0.005, 0.150 and 0.350 labelled as O, A,
B and C, respectively (see Fig.5.17). . . . . . . . . . . . . . . . . . . . . . 118
5.19 Polar diagrams of the ratio between the length d2 and its initial value d0 : at
different axial strains ε11 = 0.000, 0.005, 0.150 and 0.350 labelled as O, A,
B and C, respectively. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 119
5.20 Polar diagrams of the opening angle α: at different axial strains ε11 = 0.000,
0.005, 0.150 and 0.350 labelled as O, A, B and C, respectively (see Fig. 5.17).120
5.21 Deviatoric stress and volumetric strain with different initial opening angles
40◦ and 45◦ being considered, predicted by the updated H-model. . . . . . 121
6.1 Stress paths in p − q plane along triaxial tests and proportional strain tests
with different dilatant rates characterized by κ (dεv = (1 + 2κ)dε2 ). . . . . 126
6.2 Stress-strain responses along mixed proportional strain and triaxial loading
paths in 3D. At the transition state, the lateral stress is kept unchanged.
The loading of 100kPa-dense (κ = −0.7) as shown in Fig.6.1 is adopted here.
ε̇v = (1 + 2κ)ε̇1 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 127
6.3 simulation of the collapse of granular column using SPH particles and the
H-model: (a) initial configuration and (b) color map of the deformation in
direction 1 after collapse. . . . . . . . . . . . . . . . . . . . . . . . . . . . . 130
6.4 Aerial view of the Bas-Verger landslid in 2006 with associated localization
of the field measurements conducted on the landslide (Fressard et al. 2016). 131
A.1 PDFs for chained particles under two confining pressures: (a) 40kPa, PDFs
from the 40Kpa-dense are stretched by strain magnitude ratio 1.9; (b)
400kPa, the 400kPa-dense stretched by 3.2. . . . . . . . . . . . . . . . . . . 134
A.2 PDFs for grain loops under two confining pressures: (a) 40kPa, PDFs from
the 40Kpa-dense are stretched by strain magnitude ratio 1.9; (b) 400kPa,
the 400kPa-dense stretched by 3.2. . . . . . . . . . . . . . . . . . . . . . . 135
List of Tables
3.1 Initial void ratios e0 of prepared samples. The sample ‘100kPa-dense’ is

used in proportional strain tests) . . . . . . . . . . . . . . . . . . . . . . . 26
4.1 Initial void ratios e0 of prepared samples. . . . . . . . . . . . . . . . . . . . 58
4.2 The fitting parameters for different confining pressures. . . . . . . . . . . . 76
5.1 Parameters used in biaxial tests Veylon 2017 . . . . . . . . . . . . . . . . . 104
xvii
Chapter 1
General introduction
Contents
1.1 Motivation and objectives . . . . . . . . . . . . . . . . . . . . . . . 1
1.2 Structure of presentation . . . . . . . . . . . . . . . . . . . . . . . . 3
1.1 Motivation and objectives
Granular materials are widespread in nature, like sands, rocks, rices etc., and in indus-
try, for example, medicine powders (Jaeger, Nagel, and Behringer 1996). Despite their
prevalence, they present complex behavior even in the lab. Even though significant efforts
have been devoted into understanding the mechanisms that govern deformation in granular
materials subjected to compression and shear based on experiments and simulations, we
still have a blurred vision about how the interactions of contacts and grains led to complex
behavior in the system of granular materials (Papadopoulos et al. 2016).
Multiscale modeling of granular materials has received an increasing attention in the re-
cent years based on notable progress in instrumentation technology and computing power.
Multiscale modeling describes processes that aim to simulate continuum-scale behavior us-
ing information obtained from computational models of finer scales in the system, rather
than resorting to empirical constitutive models, provided that the fine-scale physics and
structure are better understood than those at a coarser scale (Fish, Wagner, and Keten
2021). Even though this modelling method cannot replace conventional phenomenological
paradigm currently in engineering which describes material behaviour directly at a coarser
scale relevant for analysis and design and easier to compare to available experimental data
rather than on microscale physics, it is believed that in future multiscale modelling could
play a critical role, especially for large deformation problems (Liang and Zhao 2019).
1
2 Chapter 1. General introduction
In the macroscopic behavior in granular materials, one significant and attractive ob-
servation is the existence of stationary state where a multitude of conformations (local
arrangements of particle structures) and conformational transitions exist for which the
macroscopic stress and porosity remain constant under continuous shearing. Such macro-
scopic states is named as critical state in geomechanics. Critical state (CS), initially
observed in geomaterials (e.g. soils) under biaxial or triaxial loading, form the focus of
Critical State Theory (CST) (Casagrande 1936; Roscoe, Schofield, and Wroth 1958; Been,
Jefferies, and Hachey 1991) which stands as a cornerstone of constitutive theory for gran-
ular materials. Constitutive laws built on CST are however phenomenological and cannot
account for the underlying conformational transitions which are responsible for the emer-
gence of critical states. To this respect, only multiscale models may capture the microscale
detailed balance (Sun et al. 2021) responsible for critical state. But, many aspects of these
transitions remain poorly understood, and no multiscale model exists in which the mini-
mal set of conformational transitions responsible for critical state are embedded. Indeed,
novel structures, transitions and regimes are still being uncovered in granular materials
under shear, mostly thanks to the use of Discrete Element Modelling (DEM), as well as
the development of lab equipment (Rocks, Liu, and Katifori 2021).
Thus, it is intriguing to revisit critical state not only to improve fundamental under-
standing of stationary regimes that comprise large numbers of interacting units in complex
systems (Gardner and Ashby 1970; May 1972; Papadopoulos et al. 2018), but also to find
a method that can make the critical state as an emerging property in constitutive models.
This thesis focuses on understanding the mechanical responses of granular materials

under shearing, with an emphasis on the micromechanics of critical state and on the emer-
gence of CS in multiscale modelling based on a micromechanical-based constitutive model,
following several questions:
• Can we observe the critical state in different loading paths, especially the dilatant
or contractant proportional strain loading along which the volume never reaches a
constant value?
• What is the micromechanical mechanism under the stationary state?
• How can we make the critical state emerge in micromechanics-based constitutive

models without introducing empirical laws?
1.2. Structure of presentation 3
1.2 Structure of presentation
This thesis summarizes three years of research that led to two journal publications (JMPS
and PRE) and three conference proceedings (Alert materials workshop 2019, 2021, and
Powder and Grains 2021) at the meeting point of geomechanics and statistical physics.
Chapter 2 summarizes some of the relevant background literature. The discussion

mainly revolves shearing behavior in granular materials, starting from the classically continuum-
mechanical understanding and the description of the main features that are the most im-
portant to be considered in the development of a constitutive model for granular materials,
and ending at the more recent explanation from the micromechanical perspective.
Chapter 3 offers a fresh perspective on the classic concept of critical state (CS) in granu-
lar materials by suggesting that CS can be defined through the use of a single proportional
strain test. A comparison between proportional strain tests and biaxial tests simulated
with DEM is presented. The mechanical responses and fabric metrics evolve along dila-
tant proportional strain loading paths according to similar values after the strain level
has become large enough to wipe out the material memory in the homogeneous domains
considered in this analysis, i.e., the shear band area in dense samples and the whole area in
loose samples. This suggests that the micro-structure of a granular material subjected to
proportional strain loading evolves with a constant ability to withstand shearing without
volume change. Therefore, the CS concept can be generalized to a wide class of loading
paths which shows that CS acts as a general attractor irrespective of the loading path
considered (Deng et al. 2021c).
Chapter 4 examines the critical state in granular materials from a dynamical systems
theory perspective. The lifespan and life expectancy of specific cluster conformations are
tracked and quantified. Results suggest that these conformational clusters reorganize at
similar rates in the critical state regime, depending on strain magnitudes and confining
pressures. This rate of reorganization is quantified and it is found that the material mem-
ory rapidly fades, with an entirely new generation of force chains and grain loops replacing
the old within a few percent strain. The chapter extends the results of the paper (Deng
et al. 2021b).
Chapter 5 presents a novel method to make critical state emerge in a micromechanical-

based constitutive model, namely the H-model (Nicot and Darve 2011b) by introducing
a deactivation/reactivation procedure in local scale based on the results from Chapter 4.
4 Chapter 1. General introduction
The chapter starts with a review on H-model and finishes with preliminary stress-strain
results predicted from the H-model with the novel method.
In conclusion, Chapter 6 summarizes the most salient contributions and findings of this
thesis. Future lines of research related to this work are put forward.
Chapter 2
A brief literature review
Contents
2.1 Shearing behavior of granular materials . . . . . . . . . . . . . . . 5
2.1.1 Dilatancy and contractancy . . . . . . . . . . . . . . . . . . . . . . . 6
2.1.2 Softening and hardening . . . . . . . . . . . . . . . . . . . . . . . . . 8
2.1.3 Critical State . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 10
2.2 Discrete element view of granular materials . . . . . . . . . . . . 11
2.2.1 Discrete Element Method . . . . . . . . . . . . . . . . . . . . . . . . 11
2.3 Micromechanical analysis in granular materials . . . . . . . . . . 15
2.3.1 Particle/contact . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 16
2.3.2 Force chains . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 16
2.3.3 Grain loops . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 18
2.4 Multiscale modeling of granular materials . . . . . . . . . . . . . . 20
A brief literature review on the micromechanical investigations of granular materials is

presented from two different standpoints, namely geomechanics and statistical physics, in
this chapter. From a conventional geomechanical problem, the shearing behavior in granu-
lar materials is reviewed before being put into perspective with the recent micromechanical
interpretations to the conventional problem (Section 2.1). The other line of approach starts
with an introduction of Discrete Element Method (DEM) and statistical physics, and lands
with some of their applications (Section 2.2-2.3). There will be some crossover between
the two lines. In Section 2.4, a short review on multiscale modeling is presented. It may
be seen as a statistical constitutive modeling approach applied to geomaterials.
2.1 Shearing behavior of granular materials
Granular materials under shearing exhibit complex behavior because of the collective re-
arrangement of particles. It is the discrete nature of granular materials that makes their
5
6 Chapter 2. A brief literature review
behavior highly nonlinear, dissipative, and intimately linked to the microstructure. In

the particular field of geomechanics, the complex behavior has been conceptualized and
detailed conventionally through standard features such as hardening/softening, dilatan-
cy/contractancy, critical state, etc., as shown in Fig.2.1 in the framework of continuous
mechanics. In addition to the features at the macro scale, the shearing behavior can be
demonstrated by indexes at the micro/meso scale with a pair or a cluster of grains or
contacts and their collective descriptors. Enormous efforts have been devoted to observing
and interpreting these phenomena from macro to micro, from experiments to simulations,
especially in geotechnical engineering, in order to find either the mechanisms of the spe-
cific features or appropriate descriptors that can feed constitutive modeling. In short, the
aim of multiscale modeling is to narrow the gap between micro-scale characteristics and
macro-scale features. In this section, reviewing work starts from the continuum-mechanical
understanding and the description of the main features that are the most important to be
considered in the development of a constitutive model for granular materials, and ends at
the more recent explanation from the micromechanical perspective, ranging from detailed
observations to abstract theories.
Figure 2.1: Typical deviatoric stress and volumetric strain responses in a dense (left) and
a loose (right) granular assembly under a biaxial loading path.
2.1.1 Dilatancy and contractancy
Dilatancy and contractancy are one of the most important properties of granular materials,
defined by volume changes (increase and decrease, respectively) induced by shear defor-
mation. The main lineage of understanding of dilatancy in the framework of continuum
mechanics can be seen clearly thanks to the review work in Rowe 1962, Houlsby 1991, Li
and Dafalias 2000, Wan and Guo 2004 and Kruyt and Rothenburg 2016 at different ages.
The concept of “dilatancy” was initially highlighted by Reynolds in 1885 Reynolds 1885.
Dilatancy/contractancy plays an central role in constitutive modeling as it controls the
2.1. Shearing behavior of granular materials 7
non associated character of the flow rule in granular materials.
Empirically, the relationship of dilatancy with stress ratio, density (void ratio) and
material internal state has been established in a list of theories, based on a series of exper-
iments and motivated by soil behavior modeling. Taylor 1948 pointed out a link between
shear strength and dilatancy properties of sands based on a hypothesis about the energy
dissipation in frictional soils. The dilatancy equation used in Cam-Clay model by Schofield
and Wroth 1968 for soil behavior is a direct analogy of Taylor's work hypothesis. Alterna-
tively, Rowe 1962 proposed the stress-dilatancy theory which basically quantifies the effect
of geometrical interlocking of the particles on the stress state. Following Rowe's theory,
Bolton 1986 gave a very simple empirical fit to friction and dilation angles according to a
particularly comprehensive review of the experimental data. The dependency of dilatancy
on initial stresses and density has also been exhibited experimentally and theoretically in
literature (Rowe 1962; Wroth and Bassett 1965; Matsuoka 1974). It is expected that the
denser the sand, the more it will tend to expand, but at a given density, the angle of friction
reduces slightly with increasing stress level.
A physical estimation of dilatancy was initially conducted by Rowe 1962 based on

uniform disks or rods in triangular close-packed arrangements, then on regular packing
hexagons (Li and Dafalias 2000), which provides the physical basis of the relationship
between the dilatancy and stress ratio. More recently, dilatancy has been treated as the
emergence of a collective property within disordered packings. Numerical studies (Peyneau
and Roux 2008; Azema, Radjai, and Roux 2015) and experiments (Clavaud et al. 2017)
have indicated the absence of dilatancy in frictionless systems, which is different to the
previous idea about "rearrangement" that dilatancy arises purely from geometric exclu-
sive effect of hard particles. A work from Babu et al. 2020 investigated this paradox
by comparing the dilatancy effect between amorphous and lattice assemblies, and estab-
lished conditions under which dilatancy emerges naturally in frictionless sphere assemblies
through jamming-unjamming processes.
In addition, recent experiments and DEM simulations of specimens suggest that sig-
nificant levels of dilatancy emerge from highly coordinated particles motions, particularly
within shear bands (Oda and Kazama 1998; Tordesillas, Shi, and Tshaikiwsky 2011); a mi-
cromechanical study of dilatancy conducted by Kruyt and Rothenburg (Kruyt and Rothen-
burg 2016) identified two microstructural mechanisms: dilatancy due to the deformation of
mesostructures and dilatancy owing to topological changes in the contact network mainly
governed by the parameter contact anisotropy and coordination number, respectively.
There are a number of constitutive models for granular materials that incorporate phys-
ical ideas about the evolution of granular microstructure within the continuum mechanics
framework and accurately describe dilation properties. Wan and Guo 2004 bypassed the
minimization of the energy ratio and highlighted the importance of stress dilatancy and its
microstructural dependence upon the behavior of sand along various stress/strain paths
in axial symmetry conditions. Following the work of Wan and Guo 2004; Tordesillas, Shi,
and Tshaikiwsky 2011 characterized the evolution of stress-dilatancy with respect to the
evolution of mesostructures and proposed a model incorporating not only fabric but also
fabric evolution. Additional work in this direction has been carried out by Li and Dafalias
2012 and Gao et al. 2014, where the link between dilatancy and microstructure is embedded
within the flow rule in the constitutive model.
2.1.2 Softening and hardening
Strain softening and hardening are concepts originally developed for metals in plasticity
theory. During plastic deformation the loading surface is said to harden (the yield limit
increases in the stress space) or soften (the yield limit decreases in the stress space). In
triaxial or biaxial compression tests of granular materials, strain hardening and soften-
ing phenomena are exhibited as the increase and decrease in deviatoric stress, coming
along with dilatancy or contractancy normally demonstrated by the incremental volumet-
ric strain. Particularly, the strain softening is a complex behavior, characterized by a
gradual loss of shear resistance with strain after a peak strength has been reached (Prévost
and Höeg 1975; Read and Hegemier 1984). Discussions on weather the strain softening is
an intrinsic property of granular materials or a structure response have never stopped.
On the one hand, it has been viewed as an inherently material property of granular
materials and routinely contained in constitutive models, since it is commonly reported in
the experiments of dense materials (Chu, Lo, and Lee 1992; Verdugo and Ishihara 1996).
For example, Drucker, Gibson, and Henkel 1957 introduced the concept of soil as a work-
hardening material in the extended Mohr-Coulomb model. A link between current peak
strength and a state parameter which is a combination of volumetric and mean effective
stress information has been demonstrated to reproduce strain hardening in Wood and
Belkheir 1994. Yao, Hou, and Zhou 2009 proposed a unified hardening parameter that is
independent of stress paths in order to incorporate strain hardening and softening based
on the modified Cam-Clay model. This hardening parameter is used in constitutive models
for sands (Yao, Liu, and Luo 2016; Yao et al. 2019).
On the other hand, it has been argued that the observed strain softening in soils is
2.1. Shearing behavior of granular materials 9
not a material property but mainly the consequence of inhomogeneity in the deformation
field of the specimens during the loading, as pointed out by Read and Hegemier 1984 after
reviewing an extensive set of experiments to discuss the strain softening for rock, soil and
concrete. It has been demonstrated that the strain softening is strongly affected by the size
and shape of the sample prepared, and is normally accompanied by strain localization (Fu
and Dafalias 2011; Zhu et al. 2016). Also, some researchers advocate that both material
softening and structural softening exist but appear under different conditions (Sterpi 1999;
Liu et al. 2020a). For example, softening along with strain localization in biaxial tests is
regarded as structural softening, while that obtained in homogeneous samples as material
softening.
It is difficult to close the debate within the framework of plasticity theory, as experi-
ments with perfectly homogeneous material does not exist, especially in granular materials.
Various modes of bifurcation are possible and do actually develop during the experiments,
as pointed out by Hettler and Vardoulakis 1984.
More recent micromechanical investigations have helped to some extent by further un-
derstanding of the evolution of fabric and stress, and thereby by providing new insight into
strain hardening and softening (Li and Dafalias 2012; Tordesillas et al. 2012; Liu et al.
2020a). They have been characterized by some statistical physics descriptors related to
the rearrangement of grains. For instance, Liu et al. 2020a proposed mesoscopically-based
framework to interpret both hardening and softening mechanisms; Tordesillas 2007 ex-
plained the hardening and softening process by jamming-unjamming theory; to name it.
Thus, it might be asked: since strain hardening partly induced by the self-reorganization
of particles is recognized as a material property and involved into constitutive models, why
not softening that is also induced by rearrangement of particles? The hardening is the
process to be more well-organized and easier to be described than the softening process
where the structures are much more chaotic and difficult to grasp with higher heterogene-
ity within the contact network. This could be a main difficulty in describing the softening
process. In addition, it has been widely accepted that the inhomogeneity property should
be considered carefully in constitutive modeling. Anisotropy builds quickly as it relies
mostly on the contact network. If the anisotropy of contacts can be regarded as an inher-
ent property in granular materials, it would be important to incorporate strain softening
in constitutive models from a micromechanical viewpoint.
2.1.3 Critical State
Critical state (CS), initially proposed by Casagrande 1936, is a physical phenomenon ob-
served from a macromechanical perspective, in the form of a stationary state where stress
and volume tend to be constant under continuous shear strain. Dense granular materials
reach the critical state as a result of decreasing in dilatancy normally with shear band
generation, while loose materials tend to reach the same state after decreasing in contrac-
tancy. When the steady state is reached, deformation continues without volume changes.
The classic critical state theory (CST) developed by Roscoe, Schofield, and Wroth 1958;
Schofield and Wroth 1968 is the basic principle behind critical state soil models (Been and
Jefferies 1985; Been, Jefferies, and Hachey 1991; Yao, Hou, and Zhou 2009; Li and Dafalias
2015), which implies two conditions: a steady stress ratio and a steady void ratio. They
are expressed analytically by (
η = ηc = (q/p)c
(2.1)
e = ec = êc (p)
with ec = êc (p) the critical void ratio which defines the Critical State Line (CSL) in the
e − p plane.
More recently, it has been proven that not only do both stress and density converge to
a steady state at the critical state, but that this convergence and stationarity also manifest
at the micro-scale (Drescher and De Jong 1972; Fu and Dafalias 2011; Kruyt and Rothen-
burg 2014; Fu and Dafalias 2015; Kruyt and Rothenburg 2016; Zhu et al. 2016; Kawamoto
et al. 2018). For instance, Rothenburg and Kruyt 2004 developed a relation between
the critical state and coordination number based on DEM simulations; Fu and Dafalias
2011 studied the evolution of fabrics at the critical state using DEM simulations; Kruyt
and Rothenburg 2014 re-investigated the relationship between the continuum features at
the macroscale and micromechanical quantities including coordination number and fabric
anisotropy at both the particle and interparticle contact level and the grain loop scale;
Kuhn 2016b demonstrated the stationarity at the critical state as well as the evolution of
their disorder toward the critical state based on several characteristics at the microscale;
Zhu et al. 2016 investigated different mechanical features of two kinds of meso-structures
(force chains and grain loops) and proved that the two failure modes (localized and diffuse)
are homological with respect to the concept of the critical state.
According to several micromechanical investigations, it can be shown that the differ-

ence between the current porosity and the CS porosity does not in itself determine the
evolution of the micro-structure. The coaxiality between the plastic strain rate direction
and the fabric anisotropy proved to be a relevant state variable in defining critical state
(Li and Dafalias 2012; Theocharis et al. 2019). A third condition that quantifies the role
2.2. Discrete element view of granular materials 11
of fabric anisotropy in terms of its intensity and its relative orientation with respect to
loading direction should also be taken into account, in addition to the aforementioned two
conditions of constant stress ratio and void ratio in the anisotropic critical state theory
(ACST) (Li and Dafalias 2012; Theocharis et al. 2019). In ACST, a fabric-related tensor
F is adopted. To incorporate micromechanical observations into a macroscopic continuum
mechanics description, the norm F ≥ 0 and the unit-norm nF of F are considered with:
√
F = F nF , F = F : F, nF : nF = 1, trnF = 0 (2.2)
The Fabric Anisotropy Variable A is introduced by
A = F : n = F nF : n = F N (2.3)
where n represents a direction along which the loading is applied, say, the direction of
plastic flow; thus, it can be related to the stress tensor for monotonic radial loading.
N = nF : n is a measurement of the relative orientation of F and n. A tends toward 1
at critical state because both F and N tend toward 1. Then, ACST can be expressed as
follows:
η = ηc , e = ec = êc (p), A = Ac = 1 (2.4)
2.2 Discrete element view of granular materials
2.2.1 Discrete Element Method
Discrete mechanics not only helps to prove or correct the transitional understanding and
empirical theories of granular materials (as explained in the previous section), but also
brings new knowledge enriching our interpretation of the complex behavior of granular
materials under shearing. There are two types of tools used in micromechanical analysis:
experiments and numerical simulations. As for the former, the technique of particle image
velocimetry (PIV), computer tomography, magnetic resonance imaging, and confocal mi-
croscopy can provide rich microscopic information of the granular material, like the shape
of particles as shown in Fig. 2.3 (Kawamoto et al. 2018), local void ratio (), local strain
field and the geometric information of interparticle contacts and so on; in terms of me-
chanical data, photo-elastic rods are the most popular granular assemblies used to study
the force distribution and transmission within a deformed packing, as shown in Fig.2.2.
The latter refers to the discrete element method simulation is used in this PhD thesis to
run numerical experiments with unlimited access to micro data.
Figure 2.2: images of a system of photoelastic disks for an isotropically compressed state
(left) and a sheared state (right) (Majmudar and Behringer 2005)
Figure 2.3: Avatar specimen under isotropic stress (left) and at the end of triaxial loading
(right) (Kawamoto et al. 2018).
2.2. Discrete element view of granular materials 13
Discrete element method (DEM), developed by Cundall 1971; Cundall and Strack 1979,
has been widely accepted as a powerful method to simulate the complex mechanical be-
havior of granular materials at the Representative Elementary Volume (REV) scale. Note
that REV proposed by Hill 1963 is the smallest volume providing representative mechanical
quantities at macroscale, like stress and strain, in the framework of continuum mechanics.
DEM simulations lay out abundant details at the microscopic level, from geometric data
to force and energy information, which is relative difficult to be accessed in concrete ex-
periments. DEM has been regarded as a reliable tool to find the connection between local
physics and macro properties of granular materials in considerable investigations includ-
ing those reviewed in the previous section. In this subsection, Discrete Element Method
(DEM) is briefly reviewed.
All DEM simulations performed in this PhD thesis are based on the open-source DEM
software YADE (Yet Another Dynamic Engine), which is developed with C++ and Python
programming languages (Smilauer et al. 2015).
2.2.1.1 Computation loop
In DEM, as introduced in the pioneering work (Cundall and Strack 1979), the interaction
between particles is viewed as a transient problem with states of equilibrium developing
whenever the internal forces balance. The numerical scheme is explicit. The equilibrium
contact forces and displacements of a stressed assembly of non-deformable discs are found
through a series of calculations tracing the movements of the individual particles, and the
movements are the result of the propagation through the medium of disturbances origi-
nating at the boundaries: this is therefore a dynamic process. The dynamic process is
described numerically. It is assumed that the time step chosen should be small enough in
order that disturbance cannot propagate from any disc further than its immediate neigh-
bours during a single time step. The calculations performed in DEM alternate between
the application of Newton's second law to the discs and a constitutive law at the contacts.
Newton's second law gives the motion of a particle resulting from the forces acting on it.
The constitutive law is used to compute contact forces from displacements.
Considering a granular material at time t0 , geometrical data describing the positions

and size of grains and walls, and physical properties of the grains are given. To determine
the contact forces and positions of grains at time t0 + ∆t, there are four steps as follows:
• Detect the contacts: two spherical grains are in contact only if the distance between
centers of two particles is less than the sum of their radii. If this condition is met,
go to the next step.
• Compute the contact forces based on the contact law and relative displacement in-
crements.
• Integrate the force and momentum equations for each grain, and calculate the accel-
eration of them according to Newton's second law.
• Update the positions of the grains by integrating the acceleration during ∆t.
After that, the particle positions are updated at this time step, and become the initial
state of the next time step. Then, repeat the circulation under control.
2.2.1.2 Contact law
The contact law is one important modeling hypothesis in DEM simulations since it af-
fects the macroscopic behavior significantly together with particle arrangement. In DEM,
some contact laws use strains and stresses while others are expressed through displace-
ment - force relation. The contact law presented here is the most common and classic
one in DEM, originally proposed by Cundall and Strack 1979 as the simplest non-cohesive
elastic-frictional contact model. As shown in Fig.2.4, when a new contact is established,
it is assumed that there are two linear springs awakened: the one in the normal direction
and the other in the tangential direction to the contact, characterized by the normal and
tangential stiffness kn and kt respectively. A Coulomb-type friction law is incorporated by
introducing the intergranular friction angle ϕ.
Consequently, the normal contact force Fn is defined by the overlapping distance un be-
tween the two particles and the normal contact stiffness kn , expressed as Equation 2.5;
the tangential contact force Ft after being assessed by the friction law can be written as
Equation 2.6. Note that the stiffness kn depends on a material stiffness E, and the radii
of the two particles R1 and R2 , and kt is determined from kn and a stiffness ratio α.
Fn = kn un (2.5)
(
kt ut if |Ft | < Fn tanϕ
Ft = (2.6)
tanϕFn otherwise
2R1 R2
kn = E (2.7)
R1 + R2
2.3. Micromechanical analysis in granular materials 15
Figure 2.4: Elasto-frictional contact law: components in the contact (left), behavior in the
tangential direction (right).
2.3 Micromechanical analysis in granular materials
As assumed in DEM simulations, Granular materials can be regarded as assemblies of rigid

particles, where particles interact through contact forces and small contact areas. Thus, the
microscale level is that of particles and contacts (Kruyt and Rothenburg 2014), as shown in
the Fig.2.4. Granular materials can be decomposed into weak network and strong network
according to the force transmission, or grain loops tessellating the 2D assembly, and many
other types of mesoscopic cluster. In summary, the structural object of granular materials
can be grouped through different scales: (i) microscale at the particle and contact level,
(ii) mesoscale involving a cluster of grains, (iii) macroscale of the global contact network.
Related to these diverse structures, variables can be constructed from various aspects
(Kuhn 2016b). Information of the state (being) and transition (becoming) can be involved
in to the following three perspectives: (i) configuration quantities, such as coordinating
number, contact orientation; (ii) force quantities, such as particle or cluster stresses and
contact forces; (iii) movement quantities, such as particle translations, local strains. The
first two aspects are spacial variables associated with the status of granular assembly. The
third set of attributes is space and time related, comprising transitional or rotational rate
quantities that are driven by the bulk deformation and depending on the loading direction.
2.3.1 Particle/contact
The microscale level in granular materials is that of particles and contacts. The relative
displacement and the contact force are information of interest (Kruyt and Rothenburg
2014; Imole et al. 2014). The coordination number, the fabric tensor and the distribution
function of orientations of contacts or particles etc. are the most popular statistics based
on this information used in multiple analysis.
The coordination number is the average number of contacts per particle, and refers to
the compact degree (closely linked to the density and porosity) of the material (Kruyt and
Rothenburg 2014). It is defined as
2Nc
Z= (2.8)
Np
where Nc and Np represent the number of contacts and particles, respectively.
Fabric tensor collects the orientation information in contacts or particles. Different types
of fabric tensors can be constructed with different vectors (Oda 1972; Satake et al. 1982;
Fu and Dafalias 2011). For example, a fabric tensor of contact normals nci can be expressed
by
1 X c c
Fij = nn (2.9)
Nc c∈C i j
where the sum is over all contacts c in the set of contacts C. Similar quantities can be
constructed with the direction of contact force for instance.
2.3.2 Force chains
Force chains, a concept of quasilinear particle clusters where stress is concentrated, have at-
tracted widespread interest. There is abundant experimental and numerical evidence that
the spacial distribution and temporal evolution of the force structures on the mesoscale
govern macroscopic behavior. Particularly, the static state in the sample and the buckling
(or bending) under loading of force chains play the primary role in carrying external load
and energy dissipation, respectively (Drescher and De Jong 1972; Radjai et al. 1996; Peters
et al. 2005; Tordesillas 2007; Tordesillas, Zhang, and Behringer 2009; Tordesillas, Walker,
and Lin 2010; Saitoh, Magnanimo, and Luding 2015; Wautier, Bonelli, and Nicot 2018a).
An objective description of what constitutes a force chain initially proposed by Peters

et al. 2005 is illustrated in Figure 2.5 and summarized as follows:
• The number of contacting particles in a force chain is not less than 3.

Figure 2.5: Force chain definition (Peters et al. 2005).
• The major principal stress on each particle belonging to a force chain exceeds the
global average value.
• The major principal stress direction is aligned with the contact direction, for example,
with less than 45◦ deviation.
The stress tensor on each particle is computed according to the summation running
over all external forces of the particle, as shown in Equation 2.10 in the initial method
(Peters et al. 2005). A modified approach proposed in Muthuswamy and Tordesillas 2006
adopted a force moment tensor replacing the stress tensor for each particle calculated as
in Equation 2.11 to avoid preference for small particles.
N
1 X c c
σij = f r (2.10)
V c=1 i j
where V is the volume of the particle, N is the number of contacts of the particle, fic is the
ith component of the force acting at the contact, and rjc the jth component of the radius
vector from the center of the particle to the point of contact.
N
1X c c
Sij = f r (2.11)
r c=1 i j
where r is the particle radius.
In short, there are three thresholds needed to detect a force chain: magnitude of the
principal stress (or force), direction of the principal stress (or force), and the minimum
number of particles.
As for the force chain itself formed in a specimen, it is affected by multiple factors. Find-
ings from Muthuswamy and Tordesillas 2006 indicate that increasing interparticle friction,
packing density and degree of polydispersity promotes the formation of straighter chains
and a greater degree of branching in the force chain network. Denser systems tend to favour
shorter chains. Thus, straighter and shorter force chains, combined with a greater degree of
branching in the force chain network, result in a macroscopically stronger granular material.
The failure mode of force chains was investigated numerically and experimentally by
Tordesillas, Zhang, and Behringer 2009. It is demonstrated that frictional rolling is the
predominant mode of contact failure in a force chain buckling, and in the contacts with
and within its laterally supporting neighbours.
Most importantly, connections of force chains of discrete medium to the formation of

the most compressive principal stress, and of force chains buckling to energy dissipation,
unjamming transition and global failure have also been looked into carefully in literature
(Oda and Kazama 1998; Radjai et al. 1998; Tordesillas 2007; Tordesillas, Zhang, and
Behringer 2009; Tordesillas, Shi, and Tshaikiwsky 2011; Luding, Jiang, and Liu 2021).
Force chain buckling has been regarded as the characteristic of unjamming events at the
mesoscale, to be responsible to non-affine strain, and the mechanism for energy dissipation
(Tordesillas 2007). It has been attributed the governing mechanism for shear banding to
force chain buckling that is concentrated inside shear bands, and responsible for the high
levels of dilatancy observed within these bands (Tordesillas 2007; Tordesillas, Zhang, and
Behringer 2009; Tordesillas, Shi, and Tshaikiwsky 2011).
2.3.3 Grain loops
Grains loops initially attracted attention for computing local strain (Satake 1992; Kruyt
and Rothenburg 1996; Kuhn 1999; Kruyt 2003). The geometric structure of a 2D granular
assembly can be described by a particle graph which entirely subdivides the assembly into
polygonal sub-domains, namely grain loops (also known as mesoloops or mesodomains), as
shown in Figure 2.6. Thus, grain loops, consisting of contacting grains, are an important
type of mesostructure in 2D in granular materials in terms of local strains. Note that force
cycles mentioned once in the previous subsection is an example of the grain loops defined
here.
The stress and strain tensors of each grain loop are defined from the contact forces and
particle movements by (Nguyen et al. 2009; Nguyen, Magoariec, and Cambou 2012). The
elongation degree and elongation direction defined based on the loop fabric tensor have
Figure 2.6: Grain loops categorized by edge length in 2D granular assembly.
been used in Nguyen et al. 2009 to investigate the relation between structure and strain at
the mesoscale. It has been revealed that strain in grain loops significantly depends on the
elongation degree and the orientation of loops. It was shown that contractancy takes place
within the grain loops which are elongated along the minor principal compression direction
whereas dilatancy takes place within the grain loops which are elongated along the major
principal compression orientation. In addition, the structure and stress on grain loops have
been connected by Nguyen, Magoariec, and Cambou 2012. It is demonstrated that the
distribution of the local mean stress is uniform in any grain loop, while the distribution
of the deviatoric stress is significantly dependent on the elongation direction and on the
elongation degree of the grain loops. The local stress ratio is higher within the grain loops
that are elongated in the global compression direction than that within the ones elongated
in the global extension direction.
In addition, grain loops at the critical state where the specimen volume keeps constant
under shearing have been investigated by Zhu et al. 2016; Zhu, Nicot, and Darve 2016;
Kruyt and Rothenburg 2014. The results from Zhu et al. 2016; Zhu, Nicot, and Darve
2016 demonstrate that the percentage of each loop category (grouped according to the
edge number) reaches a steady state that corresponds to the critical state regime at the
specimen scale. Consistent to this finding, the results on the topological exchange within
grain loops along a biaxial test show that difference between the new and disappearing
loops follow a non-monotonic evolution, until fluctuating around zero at the critical state
for the different categories of grain loops (Liu, Nicot, and Zhou 2018). At a specific state
belonging to the critical regime, the distributions of loop orientation characterized by the
loop fabric tensor are identical for each category independent of the coordination number
(Kruyt and Rothenburg 2014). The interplay between force chains and force cycles (N
cycles whose contacts each bear above average force) has been examined by Tordesillas,
Walker, and Lin 2010. It is unraveled that the three-force cycles play an primary role in
reinforcing the force chains and indicating failure resulting from force chain buckling. The
significant role of grain loops in supporting force chains in the framework of force networks
is another reason for considering grain loops (normally referred to as force cycles instead)
(Tordesillas, Walker, and Lin 2010).
2.4 Multiscale modeling of granular materials
A multiscale modeling of granular materials involves a trade-off between precision and

simplicity in describing the microstructure. One should consider at which scale to simplify
the physical information because of the existence of multiple spatial and temporal scales.
It follows that studying structures at multiple length scales offers a fruitful pathway for
the understanding of static and dynamic properties of granular materials. Physical infor-
mation can be incorporated in different ways at different length scales.
The multiscale framework of coupling Finite Element Method (FEM) and Discrete El-
ement Method (DEM) is promising method to deal with small scale problems. It takes
advantage of the efficiency of FEM at solving boundary value problems at structural level
and the capability of DEM to capture complex material behaviors (Meier, Steinmann,
and Kuhl 2009; Guo 2014; Shahin et al. 2016; Desrues et al. 2019). This approach can
reproduce interesting experimental observations such as the inception of shear band un-
der smooth symmetric boundary conditions, large dilation, critical state reached within
the shear band, etc. Nevertheless, these DEM–FEM methods have a limitation when
geotechnical-engineering problems are considered due to the fact that a great number of
particles need to be contained in a boundary value problem (BVP) in order to reach local
convergence. It makes these methods waste computational resources resulting in inefficient
simulations. In addition, all the information contained in the DEM simulation is not used
because of averaging, which makes the approach not very economical in terms of data saved
in memory.
Micromechanically-based constitutive models should be a good alternative to couple

with a numerical methods e.g., the finite element method (FEM) (Xiong, Yin, and Nicot
2019), the smooth particle hydrodynamics (SPH) method (Xiong et al. 2021), and Finite
Difference Method (FDM) (Wautier et al. 2021). There are different types of micromechan-
ical constitutive models (Li and Dafalias 2012; Chang and Hicher 2005; Nicot and Darve
2011b). Some of them incorporate statistical physics into the transitional phenomenal laws
2.4. Multiscale modeling of granular materials 21
(Li and Dafalias 2012), the other ones construct a model from geometrical features and
try to recover the mechanical behavior at macroscopic scale explicitly, for example the
H-model (Nicot and Darve 2011b).
It is now recognized that the relative displacement of any contact is not proportional
to the macroscopic strain. In the H-model (Nicot and Darve 2011b), one of most impor-
tant hypotheses is that this approach describes the local behavior through mesostructures
instead of contacts. One main advantage of such an approach is to give access to rich
microstructural information that is out of reach for standard phenomenological models in
which the microstructure is simplistically embedded as internal variables, and is not de-
scribed in an explicit way. Such approaches try to use a minimal set of data available in
DEM simulations. It is more computationally affordable than DEM.
Chapter 3
On the attraction power of critical state
Contents
3.1 DEM simulation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 24
3.2 Mechanical response . . . . . . . . . . . . . . . . . . . . . . . . . . . 27
3.2.1 Mechanical response along dilatant proportional strain loading paths 27
3.2.2 Mechanical response along biaxial loading paths . . . . . . . . . . . . 28
3.3 p-q-e space analysis . . . . . . . . . . . . . . . . . . . . . . . . . . . 30
3.3.1 p-q plane . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 30
3.3.2 p-e plane . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 38
3.4 Fabric-related critical state locus . . . . . . . . . . . . . . . . . . . 39
3.4.1 Fabric tensor analysis . . . . . . . . . . . . . . . . . . . . . . . . . . 39
3.4.2 Grain loop evolution . . . . . . . . . . . . . . . . . . . . . . . . . . . 41
3.5 Mixed proportional strain and biaxial loading paths . . . . . . . 46
3.6 Conclusion and outlook . . . . . . . . . . . . . . . . . . . . . . . . . 52
Although the structure of a granular material appears to be simple at the microscopic

scale, its behavior shows itself to be complex at the macroscopic scale, mostly due to the
collective rearrangement of particles. Based on the gain or loss of contacts, granular ma-
terials will easily adjust to any change under loading conditions. Among the accessible
microstructures, some have the ability to withstand constant shearing with no change of
volume. A state with such microstructures is known as critical state (CS) and plays a
leading role in the most popular constitutive relations for granular materials, as reviewed
in the second chapter.
Research on CS is nearly exclusively based on conventional tests (triaxial tests in 3D

or biaxial tests in 2D) in which a constant lateral stress condition dσlateral = 0 is imposed
together with a constant strain rate in one given direction dεaxial = constant (Been and
Jefferies 1985; Been, Jefferies, and Hachey 1991; Fu and Dafalias 2011; Fu and Dafalias
23
24 Chapter 3. On the attraction power of critical state
2015; Zhu et al. 2016). This preferential selection is linked to the tendency of generat-
ing constitutive models to work with stress and to calculate the resulting strains, as well
as the difficulty of imposing fully controlled strain paths in the laboratory (Ibraim et al.
2010; Daouadji et al. 2017). For conventional loading, the p − q path is preset; sam-
ples are loaded by increasing the axial incremental strain under constant lateral confining
pressure; volumetric strain and deviatoric stress responses are recorded according to the
loading path. This test allows for the existence of stationary states in volume and stress
(the so-called critical state). Proportional strain tests, on the other hand, have been much
less studied whereby the loading is imposed by proportional strain rates in the axial and
lateral directions with dεlateral = λdεaxial and dεaxial = constant. Under proportional strain
loading condition, the volumetric strain is no longer a response but a loading variable and
the stress path is not known beforehand, which makes the interpretation of such tests
more complex than it is for triaxial or biaxial tests. Except for the particular case of the
undrained triaxial test, whereby the volume is kept constant, the continuous change in
volume prevents stationary states in volume and stress along proportional strain paths to
be observed. Without the existence of a stationary state, investigating the evolution of
the microstructures along a proportional strain loading path becomes a major challenge.
In other words, tracking the relation between triaxial (or biaxial) and proportional strain
loading paths in terms of mechanical responses and underlying microstructures is an open
and stimulating topic.
In order to generalize the concept of CS, we focus on the relation between the material
responses under dilatant proportional strain loading paths and biaxial loading paths in
2D. These two types of loading paths, as well as mixed biaxial/proportional strain paths,
have been simulated by DEM. The details of the DEM simulation are shown in Section
3.1. Mechanical responses along biaxial tests and proportional strain paths are presented in
Section 3.2. The relation between proportional strain tests and biaxial tests in the p − q − e
space is analysed in Section 3.3. In Section 3.4, the fabric-related CS locus is studied. The
mechanical responses along the mixed loading paths are presented in Section 3.5, as is the
relation between the imposed dilatancy in proportional strain tests and the dilatancy in
biaxial tests.
3.1 DEM simulation
The Discrete Element Method (DEM) (Cundall and Strack 1979) is a powerful numerical
method to simulate the global behavior of a set of grains interacting through contact laws.
It has been widely used to simulate the mechanical response of granular assemblies under
3.1. DEM simulation 25
various loading conditions. The mechanical states are characterized locally by kinematic
information including position, rotation and velocity of grains, as well as static informa-
tion based on contact force between contacting particles. The open source software YADE
(Smilauer et al. 2015) has been used.
Under consideration here is a quasi-2D soil sample in the form of an assembly of a

single layer of 20,000 spherical particles contained within a surface domain of 1 m × 1.5 m.
The particle sizes have an average diameter d50 = 0.008 m and dmax /dmin = 2. The elasto-
frictional law introduced by Cundall and Strack 1979 has been adopted as the contact law.
The cohesiveless contact parameters between two grains contain a normal and tangential
linear spring of respective stiffness kn and kt , as well as a friction characterized by a friction
angle ϕ = 35◦ . kn /Ds is given as 300 MPa, where Ds = 2R1 R2 /(R1 + R2 ) and R1 , R2 are
the radii of particles in a given contact and kt /kn is 0.5. If the pressure of 100 kPa is con-
sidered, the average ratio of contact overlap and particle size < un > /d50 is around 0.16%,
which is close to the value (10−3 ) in literature (Kruyt and Rothenburg 2016; Theocharis
et al. 2019). For sample preparation, two schemes can be used to obtain an isotropic
compression of the specimen: the boundary moving scheme and the internal compacting
scheme (Wautier, Bonelli, and Nicot 2019a). In this study, particles are enlarged, at first,
while keeping the boundary walls fixed to generate a preliminary sample up to 90 kPa.
Then, to achieve a more precise consolidation pressure, the sample undergoes an isotropic
consolidation by imposing an equivalent incremental strain on the boundaries in the ver-
tical and lateral directions up to a specific consolidation state. During this preparation
process, contact friction angles of 2◦ and 35◦ are used to obtain dense and loose samples,
respectively. Consolidation pressures of 10 kPa, 20 kPa, 40 kPa, 60 kPa and 100 kPa are
considered. Three dense samples and five loose samples are thus prepared and named as
indicated in Table 3.1. The void ratio is accounted for, based on solid and void surfaces
on the 2D plane as shown in Fig.3.1. During the shearing process, a contact friction angle
ϕ = 35◦ is adopted in all the samples. Note that the soil mechanics convention is adopted
throughout the thesis with compression and contraction counted positive. As a result,
dilatant/contracting volumetric strain are considered as negative/positive respectively.
For proportional strain tests, 100kPa-dense sample is used with an initial void ratio
of 0.191. In the 2D simulation, dε1 = λdε2 and dε2 = 10−2 s−1 . As dεv = (λ + 1)dε2 ,
λ > −1 corresponds to a contracting test, λ = −1 to a constant volume test, and λ < −1
to a dilatant test. We have adopted λ = −1.2, −1.3, −1.4 for the three dilatant propor-
tional strain tests. Biaxial loading paths with corresponding pressures are applied to all
the samples indicated in Table 3.1. The stress and strain states are then described in 2D
as follows: deviatoric stress q = σ2 − σ1 , mean stress p = (σ1 + σ2 )/2, and volumetric strain
εv = ε1 + ε2 , where σ1 and σ2 are the principal stresses at horizontal and vertical directions
and ε1 and ε2 are the principal strains. The direction 2 is always the vertical loading
control direction (ε̇2 = 10−2 s−1 ), as shown in Fig.3.1. When the pressure ranges from 20
kPa to 200 kPa, the Inertial Number I of the granular system ranges among 2.77 × 10−5 -
8.76 × 10−5 , which corresponds to a quasi-static state (Da Cruz et al. 2005).
In reality a 2D model of a granular material is quantitatively different from sand as-

semblies; for example void ratio values and coordination numbers in 2D assemblies are
smaller than those in 3D (Fu and Dafalias 2011), and stresses expressed in kPa relies on
the use of an arbitrary out of plane dimension 1 . However, given that the complexity of
the constitutive behaviour of granular assemblies stems mainly from the local properties
and the disordered packing, both effects can be captured in 2D simulations (Nicot and
Darve 2011b). Qualitative investigations of granular materials based on DEM simulations
in 2D can then be considered as both effective and efficient. Proof of such effectiveness
and efficiency can be found in various studies of the microscopic mechanism behind the
mechanical responses of granular materials under different loading paths have been inves-
tigated (Fu and Dafalias 2011; Kruyt and Rothenburg 2014; Fu and Dafalias 2015; Zhu
et al. 2016; Liu et al. 2020a). Thus, the 2D assembly simulation is adequate enough for
qualitative investigations of critical states under proportional strain and biaxial loadings
performed in this study.
Table 3.1: Initial void ratios e0 of prepared samples. The sample ‘100kPa-dense’ is used
in proportional strain tests)
Sample 40kPa-dense 60kPa-dense 100kPa-dense 10kPa-loose

e0 0.197 0.195 0.191 0.266
Sample 40kPa-loose 60kPa-loose 100kPa-loose 20kPa-loose
e0 0.267 0.265 0.261 0.267
1
In the present study, the out of plane dimension is then taken equal to the thickness of the sample
box (0.04 m), which allows kPa rather than kN/m to be used for the stress unit on the boundary walls.
The thickness of the box has a proportional influence on the stress values, which should not impact the
following results, at least qualitatively.
3.2. Mechanical response 27
(a) (b)
Figure 3.1: Quasi-2D DEM specimens for (a) proportional strain tests where loading pa-
rameters are ε1 and ε2 and (b) biaxial tests where loading parameters are ε2 and σ2 . The
direction 2 is always the vertical loading control direction at the REV scale.
3.2 Mechanical response
3.2.1 Mechanical response along dilatant proportional strain load-

ing paths
3.2.1.1 Stress-strain analysis
Figure 3.2 shows the evolution of the deviatoric stress q along the axial strain ε2 under
dilatant proportional strain loading path with λ = −1.2, −1.3, −1.4. In this figure, it can
be observed that the deviatoric stress q grows quickly at first and then decreases after the
peak. Gradually, q decreases to 0. When λ = −1.4, the sample undergoes the fastest
dilatancy and the corresponding stress curve reaches the earliest peak at ε2 = 0.013 and
the earliest zero deviatoric stress at ε2 = 0.061. When λ = −1.3 and λ = −1.2, it appears
that the peaks are reached at ε2 = 0.017 and ε2 = 0.019 because of smoother and smoother
dilatancy. Zero mean pressure p occurs at ε2 = 0.085 and ε2 = 0.142, respectively.
3.2.1.2 Kinematic pattern: from localization to diffuse
The existence of a stress peak (Fig.3.2) corresponds to a generalized limit state (Nicot
and Darve 2011a) which has been shown to be a proper failure state (Wan, Nicot, and
Darve 2017). Various failure modes, characterized by localized or diffuse patterns, can be
encountered after the peak, along the descending branch. In biaxial tests, either diffuse
or localized modes will appear (Nicot and Darve 2011a). In this study, however, these
two modes appear successively along each proportional strain test. In this subsection, the
incremental deviatoric strain distribution has been used to characterize the kinematic pat-
tern, as introduced in literature (O’Sullivan, D. Bray, and Li 2003; Zhu et al. 2016). For
example, in the test with λ = −1.2, a typical diagonal shear band traversing the whole
specimen appears at ε2 = 0.008, as shown in Fig.3.3 (I), corresponding to the point I in
Fig.3.2. After a process of dilatancy, the shear band begins to vanish in the strain state
ε2 = 0.110 shown in Fig.3.3 (II). Fig.3.3 (III) demonstrates the kinematic pattern when
ε2 = 0.142, where the mean pressure p is close to zero. An evolution from localization,
when the strain largely concentrates in a partial domain of the material, to liquefaction,
when the effective stress within the granular specimen is reduced to essentially zero, has
been observed along the three dilatant proportional strain paths, as shown in Fig.3.3. With
the increase in volume along the loading path (λ < −1), the density of the sample grad-
ually decreases and the kinematic pattern changes from localized to diffuse until a zero
mean pressure p is reached.
Kinematic patterns characterize the nature of the failure mode, together with the spatial
domain. It has been widely accepted that fabric-related measures should be taken either
within the shear band or within the full sample when the shear deformation is diffuse, since
measures within nonhomogeneous domains have no constitutive meaning (Desrues et al.
1996; Fu and Dafalias 2011; Fu and Dafalias 2015; Zhu et al. 2016; Salvatore et al. 2017).
In this study, the index Med (Liu, Nicot, and Zhou 2018) has been adopted to define the
shear band domain. Med refers to an absolute difference of incremental deviatoric strain
inside and outside the shear band with a trial width divided by their sum. Regarded as an
optimization problem, the shear band width has been obtained according to the maximum
Med within a reasonable range of the trial shear band width. More details are available
in Liu, Nicot, and Zhou 2018. In the following section, labels with * refer to measures
within a homogeneous domain (shear band for dense specimen and whole sample for loose
specimen). Taking the kinematic evolution as its base, the next section will examine the
evolution of the stress and void ratio.
3.2.2 Mechanical response along biaxial loading paths
As recalled in the introduction, the critical state (CS) refers to a state where stresses,
void ratios and fabrics tend to be steady at relatively large deformation when shear strain
further increases (Theocharis et al. 2019). Biaxial tests used as references have been simu-
3.2. Mechanical response 29
lated to define the critical state line which refers to a collection of critical states obtained
at different confining pressures. Typical macroscopic responses in biaxial tests are illus-
trated in Fig.3.4, where deviatoric stress and volumetric strain evolve with respect to axial
strain ε2 . Figures on the left hand show the results from the dense samples. In these three
figures, the deviatoric stresses show an increase before a peak is reached, and a decrease
followed by a steady regime with small fluctuations. The stress peaks rise gradually with
the increase in confining pressure from 40 kPa to 100 kPa. The volumetric strain shows a
small contractancy, at first, and transfers to dilatancy before a steady state is reached. On
the other hand, for the loose specimens, q and εv continuously rise up to steady states, as
shown on the right-hand panel of Fig.3.4. Localized and diffuse patterns have developed
within the dense and loose samples, respectively.
As highlighted in Subsection (3.2.1.2), it is important to focus on the domain within

the shear band when considering void ratio and fabric indexes in order to characterize ma-
terial scale properties by considering only homogeneous domains. The same point holds for
stresses even if the stress heterogeneity is never reported in strain localization problems.
Do stress patterns exhibit significant differences inside and outside the shear band? Is
the stress path within the shear band the same as in the whole sample? Usually, stresses
are obtained macroscopically from the external forces applied on the sample boundaries.
To obtain the stresses within shear band, we have adopted a local definition based on
Love-Weber stress at the grain scale. The mean stress tensor σ̄ijp of each particle can be
computed based on the contact forces applied, thanks to the Gauss theorem (Nicot et al.
2013b; Liu et al. 2020b). More details are available in aforementioned literature. The stress
tensor for a given domain Ω can be computed based on all Np particles in the domain as
Np
σij∗ = Ω1 σ̄ijp V p .
P
p=1
In Figure 3.5, global stress components have been compared with local stresses com-
puted within the shear band domain for the 100kPa-dense sample under a biaxial loading
path. Stresses based on boundaries, including principal stresses σ1 and σ2 in the whole
sample, are compared with σ1∗ and σ2∗ considering the domain of the shear band based on
the grain scale stress definition. In addition, the relative orientation of the stress tensor
characterized by (θσ1∗ − θσ1 ) is presented. It is observed that the evolution of stresses within
the shear band and the whole sample is far more similar than it is for the void ratio (Zhu
et al. 2016) in 2D. Thus, it is reasonable to adopt stresses based on sample boundaries
instead of stresses within the shear band when investigating critical state. This approach
has been adopted throughout the thesis. A precise characterization of the loading path
within the shear band and its difference from the whole specimen is an issue that remains
to be examined. Some preliminary results on stress rotation within shear bands can be
found in (Liu et al. 2020a) and more investigations and discussions on this issue will be
100kPa-dense ( = 1.2)
120 I 100kPa-dense ( = 1.3)
100kPa-dense ( = 1.4)
100
80
q (kPa)
60
40
20 II
III
0
0 2 4 6 8 10 12 14 16
2 (%)
Figure 3.2: Evolution of the deviatoric stress q along dilatant proportional strain tests with
different magnitudes of dilatancy characterized by λ = −1.2, −1.3, −1.4. Point I, point II
and point III refer to onset of a well marked shear band at ε2 = 0.010, slightly blurred
shear band at ε2 = 0.109, and the pressure p close to zero at ε2 = 0.142, respectively. The
corresponding kinematic patterns of these three points are shown in Fig.3.3
conducted in further studies.
3.3 p-q-e space analysis
The CSL is characterized by mean stress p, deviatoric stress q and void ratio e, usually
plotted separately in planes (p, q) and (p, e). So, here, mechanical responses (p, q, e) from
proportional strain tests and biaxial tests have been compared in p − q and p − e planes.
3.3.1 p-q plane
Figure 3.6 illustrates the stress paths of the biaxial tests and the proportional strain tests.
The three straight lines result from biaxial tests with dense samples. The initial states
are marked with triangles and ultimate points are highlighted by squares labelled as A, B
and C. The critical stress ratio line (CSRL) is given as a dashed line based on the three
3.3. p-q-e space analysis 31
(I) (II) (III)

Figure 3.3: Incremental deviatoric strain maps (dεd ) estimated for axial strain increments
of 0.11%. Three axial strains are considered along the proportional strain path λ = −1.2:
(I) onset of a well marked shear band at ε2 = 0.010, (II) slightly blurred shear band at
ε2 = 0.109, (III) just before liquefaction (p is close to 0) at ε2 = 0.142. The corresponding
points in q − ε2 plane are shown in Fig.3.2
ultimate points, and the three highest points define the maximum stress ratio line (MSRL).
As for the dilatant proportional strain tests, Fig.3.6 shows that the path of the devia-
toric stress and the mean stress (p, q) for each case, starts from the initial state of around
(100 kPa, 0), evolves to the MSRL and, at the end, changes direction to turn back toward
the zero-stress level along an asymptotic line after having formed a ‘loop’. Result of a
mixed biaxial test/proportional strain test is also presented in Fig.3.6. The sample was
led to CS by a biaxial loading under 100kPa lateral stress upon which the loading mode
was switched to a dilatant proportional strain loading. It can be seen that the stress path
evolves from the very beginning along the CSRL with decreasing p and q.
Figure 3.5 illustrates that the stresses in the sample are relatively homogeneous at least
in a 2D condition. The vanishing of the principal stresses along the dilatant proportional
tests occurs simultaneously with the evolution from localized to diffuse failure (Figure 3.3).
The vanishing of the principal stresses offers two possibilities: (1) liquefaction occurs only
within the shear band and the behavior outside the shear band is regarded to be close to an
elastic solid under unloading; (2) the kinematic pattern evolves gradually to become diffuse
throughout the whole sample. In addition, it is worth noting that all computed results
display an asymptotic behavior in the p − q plane, approaching the CSRL independently
of the imposed dilatancy ratio, as shown in Fig.3.6. This kind of stress track, especially
the loop, has scarcely been reported in the literature for dilatant proportional strain tests
in either laboratory (Ibraim et al. 2010; Sibille et al. 2015; Daouadji et al. 2013) or DEM
simulations (Nicot, Sibille, and Hicher 2015), with exception of the work reported by (Chu,
40kpa-dense 40kpa-loose
80 2 80 2
60 1 60 1
q (kPa)
q (kPa)
(%)
(%)
40 0 40 0
v
20 1 20 1
Deviatoric Stress Volumetric Strain
00 5 10 15 2 00 5 10 15 2
2 (%) 2 (%)
2 2
100 100
1 1
75 75
q (kPa)
q (kPa)
(%)
(%)
0 0
50 50
v
v
25 1 25 1
00 5 10 15 2 00 5 10 15 2
(%) 2 2 (%)
200 2 200 2
150 1 150 1
q (kPa)
q (kPa)
(%)
100 0 100 0 (%)

v
50 1 50 1
00 5 10 15 2 00 5 10 15 2
2 (%) 2 (%)
Figure 3.4: Deviatoric stress and volumetric strain of biaxial tests in dense (left panel)
and loose (right panel) samples. Three biaxial loading paths with σ0 = 40 kPa, 60 kPa
and 100 kPa are considered. Note that soil mechanics convention is adopted with positive
compression.
300
* 1 2
1 1
* *
275 1 2 2.5
250
2.0
Principal stress (kPa)
225
(°)
200 1.5
1
175
1
*
1.0
150
125
0.5
100
0.00 0.02 0.04 0.06 0.08 0.10 0.12 0.14
2
Figure 3.5: Principal stresses in the whole sample (σ1 and σ2 ) and within the shear band
(σ1∗ and σ2∗ ), and relative orientation of principal stress (θσ∗1 − θσ1 ). The 100kPa-dense
sample under biaxial loading path is adopted.
Lo, and Lee 1992; Wan and Guo 2004; Daouadji et al. 2017).
To further refine the classification of the stress response along proportional strain paths,
a variety of proportional strain paths have been additionally simulated using specimens
with different densities. The results are shown in Fig.3.7. Based on these results, the stress
response path along proportional strain tests can be categorized into four types under dif-
ferent densities and volumetric strain rates, as illustrated in Fig.3.8. Group 1 results from
one typical contracting strain path along which p and q increase continuously, and Group
2 comes from a typical dilatant proportional strain path leading to zero mean pressure p
(Daouadji et al. 2013; Nicot et al. 2013a). Group 3 represents stress paths from a dense
sample under proportional strain tests with a relatively small dilatant rate, whereas Group
4 demonstrates a stress path under a proportional strain test with a contracting rate in a
dense sample, similar in trend to the stress path under an undrained loading.
It is worth paying attention to Group 3 because of the infrequently-reported ‘stress

loop’. A dense sample with imposed dilatancy was experimentally investigated by Ibraim
et al. 2010, Chu, Lo, and Lee 1992 and Daouadji et al. 2017, and numerically by Wan and
Guo 2004 and Nicot, Sibille, and Hicher 2015. The p − q relation from Ibraim et al. 2010
along such a loading path has been drawn in Fig.3.8. The stress path was stopped early
with an axial strain of around 5%. If the test had been conducted further, with the growth
of volume and a constant number of particles, an increasing number of contacts would
probably have opened that would have resulted in a stress drop at some point. Indeed, the
complete increase and decrease in q along such a loading path reported in Daouadji et al.
2017 produced a stress path that was more like a back-and-forth line than a loop. This
may have been the case because it was a middle dense sample with a not very pronounced
softening. Both the limited shearing strain and the use of the middle dense samples could
explain why no such loops were observed, even for similar loading conditions (Nicot, Sibille,
and Hicher 2015). Such loops were, however, reported in Wan and Guo 2004; Shi and Guo
2018. The Group 3 type has also been obtained in 3D simulations not included in this
chapter which focuses on 2D simulations, is presented in the perspectives in the last chap-
ter. It would be worth to conduct a detailed 3D investigation in the future.
As for the type of Group 4, stress softening was observed both along isochoric and
contracting proportional strain loading paths. The slight decrease in the deviatoric stress
q is linked to strain localization within the sample, since shear bands were generated in
these two simulations. As underlined in Wan and Guo 2004, the slope of the stress path
under a contractant proportional strain loading depends on the volumetric strain rate.
But, according to the results in Fig.3.6, the critical stress ratio in proportional strain tests
is independent of λ when the axial strain is large enough, as shown in Fig.3.6.
200
100kPa-dense ( = 1.2)
100kPa-dense ( = 1.3)
175 100kPa-dense ( = 1.4)
CS ( = 1.2)
150 40kpa-dense ( 0 = 40kPa)
60kpa-dense ( 0 = 60kPa)
Maximum stress ratio line
125 Critical stress ratio line
q (kPa)
100
75
50
25
00 25 50 75 100 125 150 175 200

p (kPa)
Figure 3.6: Stress paths of biaxial tests and proportional strain tests in p − q plane. The
start and end points are marked by triangles and squares, respectively. The critical stress
ratio curve and the maximum stress ratio curve are drawn according to the critical states
and the maximum value from biaxial tests, respectively.
1200 300
100kpa-dense ( = 0.5)
100kpa-loose ( = 0.5)
100kpa-dense ( = 1)
1000 100kpa-loose ( = 1) 250
100kpa-dense ( = 1.2)
100kpa-loose ( = 1.2)
800 100kpa-dense( 0 = 100kpa) 200
Maximum stress ratio line
Critical stress ratio line
q (kPa)
q (kPa)
600 150
400 100
200 50
Zoomed to
00 200 400 600 800 1000 1200 1400 1600 1800 00 50 100 150 200 250 300 350 400
p (kPa) p (kPa)
Figure 3.7: Stress paths in p − q plane along six proportional strain tests and one biaxial
test. Dilatant (λ = −1.2), undrained (λ = −1) and contracting (λ = −0.5) proportional
strain paths are conducted with the samples labelled 100kPa-dense and 100kPa-loose. A
biaxial loading path is performed in the sample 100kPa-dense. The square domain is
zoomed up on the right hand side.
Figure 3.8: Four categories of stress path along proportional strain tests according to DEM
simulation results shown in Fig.3.7.
0.300
0.275
0.250
0.225
0.200
e*
0.175
100kPa-dense ( = 1.2)
0.150 100kPa-dense ( = 1.3)
100kPa-dense ( = 1.4)
0.125 CS ( = 1.2)
Proprotional strain tests fitted line
0.100 101 102
p (kPa)
(a)
0.300
0.275
0.250
0.225
0.200 40kpa-dense ( 0 = 40kPa)
e*
0.175 100kpa-dense ( 0 = 100kPa)
100kpa-loose ( 0 = 100kPa)
0.150 40kpa-loose ( 0 = 40kPa)
0.125 10kpa-loose ( 0 = 10kPa)
Proprotional strain tests fitted curve
0.100 101 102
p (kPa)
(b)
Figure 3.9: p − e∗ evolution in proportional strain (a) and biaxial (b) tests. The start
and end points are marked by triangles and squares, respectively. A fit based on a power
function is shown for proportional strain tests and repeated to compare with the results
from biaxial tests, as all dilatant proportional strain tests converge towards a master curve.
The equation of the fitted curve is e∗ = 0.2571 − 0.0227( 100 p 0.6274
) .
3.3.2 p-e plane
Void ratios e∗ within the shear band are tracked for dense samples, and global void ratios
e∗ for loose samples. The void ratio is given in 2D by e∗ = (Asb − As )/As , where Asb and
As denote, respectively, the total and solid areas of the shear band. The evolution of e∗ is
represented in Fig.3.9, with respect to mean stress p.
As for proportional strain tests shown in Fig.3.9 (a), the results converge to a master
curve in the p − e∗ plane independent of the dilatant rate characterized by λ. When the
starting point of the proportional strain loading is at critical state under a biaxial loading,
the (p, e∗ ) curve will follow the master curve from the very beginning of the proportional
strain loading.
As for biaxial tests with confining pressures of 40 kPa, 60 kPa and 100 kPa, the re-
sults illustrated in Fig.3.9 (b) are consistent with experimental observations in Desrues
et al. 1996 and DEM simulations in Zhu et al. 2016. Namely, e∗ from dense samples, after
dilatancy, evolves to meet e∗ of loose samples at the critical state. For a more precise rep-
resentation of the CSL in the (p, e∗ ) plane, two biaxial test results with confining pressures
at 10 kPa and 20 kPa have been included.
These critical states with proportional strain tests in p − e∗ plane are compared. An
interesting feature stands out: the fitting curve of p − e∗ from proportional strain tests
crawls just along the critical state curve obtained from the biaxial tests. Even though the
CS from nine biaxial tests can lead to the critical state line, the master curve obtained from
fitting a curve to a cluster of p − e∗ data in dilatant proportional strain tests, compared to
the nine CS obtained from biaxial tests, induces less uncertainty.
It is worth noting that the fitting curve is based on a power function rather than a
logarithmic function. It has been proved experimentally (Verdugo and Ishihara 1996; Li,
Dafalias, and Wang 1999) that, unlike in clay, the critical state line for granular materials
cannot be estimated by a straight line in the e − logp plane. Thus, the commonly used
CSL expression ec = A + B lnpc is generally not representative for granular materials,
where the ec and pc are the critical void ratio and the critical pressure; A and B are two
constants characterizing a given material. The relation of ec and pc for granular materials
can be expressed as ec = D − E ( pFc )G , where F is a reference pressure, usually the atmo-
spheric pressure (101 kPa) for convenience, and D, E, and G are dimensionless constants
that can be identified from experiments on a given material (Li 1997; Yin and Chang 2013).
3.4. Fabric-related critical state locus 39
In summary, from Fig.3.6 and Fig.3.9, it can be observed that the master curves (p, q)
and (p, e∗ ) from dilatant proportional strain tests agree well with the critical state lines
from biaxial tests after a sufficiently large strain level has been imposed (or immediately
if the initial state is already a critical state). Hence it is shown that the classical critical
state surface defined in p − q − e space for a given granular material can be obtained by
performing one single dilatant proportional strain test. To the best of our knowledge, no
such results have been reported in the literature. For a further analysis of this relation,
the following sections will investigate the micro/mesostructures of the samples under pro-
portional strain and biaxial tests, as well as under mixed biaxial/proportional strain tests.
3.4 Fabric-related critical state locus
Even though the critical state is most often characterized by the relation between the mean
stress p, the deviatoric stress q and the void ratio e when stress and strain rate directions
are fixed (Theocharis et al. 2019), the evolution of the fabric is also an important feature
to be added to the p − q − e space in order to stay within the framework of Anisotropic
Critical State Theory (ACST) that requires the values of the fabric to be steady.
3.4.1 Fabric tensor analysis
A fabric tensor quantifies microstructural orientation-related characteristics of the material

in a tensorial form (Fu and Dafalias 2015). In this subsection, the inter-particle contact
Nc
normal directions are characterized by the second-order fabric tensor of Fc = N1c nk ⊗nk ,
P
k=1
where Nc is the total number of inter-particle contacts in the assembly; nk is the unit vector
representing the normal direction of the kth contact. Only contacts in the shear band are
considered. The norm of the normal at contact has to be normalized by the Lode angle
in order to be unique in 3D. In the present 2D case there is no Lode angle; thus, no need
for normalization. To make sure that the CS investigated within shear bands satisfies the
third condition required by ACST as mentioned before, the orientation of the fabric tensor
has been compared with the direction of the plastic flow which can be related to the stress
tensor for monotonic radial loading (Li and Dafalias 2012). The deviatoric orientation
is less than 2.5 degrees throughout the whole biaxial test on the 100kPa-dense sample.
The variable αc∗ , referring to the difference between the two principal components of Fc∗
within the shear bands, is measured to characterize fabric anisotropy. αc∗ = 0 refers to an
isotropic fabric, whereas αc∗ = 1 corresponds to the situation that the normal direction of
0.30
100kPa-dense ( = 1.2)
100kPa-dense ( = 1.3)
0.25 100kPa-dense ( = 1.4)
CS ( = 1.2)
Proportional strain tests fitted line
0.20
0.15
*
c
0.10
0.05
0.00 0 25 50 75 100 125 150 175

p (kPa)
(a)
0.30
0.25 100kpa-dense ( 0 = 100kPa)
0.20 100kpa-loose ( 0 = 100kPa)
0.15
*
c
0.10
0.05
0.00 25 50 75 100 125 150 175

p (kPa)
(b)
Figure 3.10: p−αc∗ evolution for proportional strain (a) and biaxial (b) tests. The start and
end points are marked by triangles and squares, respectively. A linear fit is shown for the
dilatant proportional strain tests and repeated on the biaxial tests, as all the proportional
strain tests converge towards a master curve.
all inter-particle contacts investigated are exactly the same.
The evolution of αc∗ has been plotted in Fig.3.10 with respect to mean stress p. Except
for the proportional strain loading starting from the CS, all the specimens have similar
fabric characteristics in terms of variable αc∗ close to 0 at the initial states, due to the
isotropic initial states where the inter-particle contact normal direction distribution is
uniform in all the possible directions. As shown in Fig.3.10 (a), all αc∗ − p graphs from
the proportional strain tests converge towards a master curve. A linear fit is given for
the proportional strain tests and repeated in Fig.3.10 (b) to compare with the results
from the biaxial tests. The relation between proportional strain tests and biaxial tests
characterized by (p, q, e∗ ) has also been observed in the p-α∗ plane, namely, the master
curve obtained in proportional strain tests has gathered the critical states from biaxial
tests. More importantly, all specimens under biaxial and proportional strain loading paths
end up with the same anisotropy in the ultimate regime. The convergent feature at ultimate
states along different loading paths is in line with the findings reported in the literature
(Fu and Dafalias 2011; Fu and Dafalias 2015).
3.4.2 Grain loop evolution
To obtain richer information on the critical state, we have analysed grain loops related
statistics in this subsection. As illustrated in Chapter 2, grain loops, enclosed by contact
branches, are formed by tessellating the material area (Satake 1992). The side number of
the loop influences considerably its deformability. In this section, loops will be categorized
into four groups according to the side number, referred to as li (i ∈ [3, 4, 5, 6p]), where ‘6p’
refers to a side number equal to or greater than 6. The percentages of categories are given
by ri = ni /nt , where ni is the population of the grain loop with a side number equal to i,
and nt refers to the total population of loops. It is worth noting that the reason for catego-
rizing r6p is that r7p (percentage of categories with a side number equal or greater than 7)
evolves in a similar way as r6 (Zhu et al. 2016).The evolution of ri∗ (i ∈ [3, 4, 5, 6p]) within
the shear band for dense specimens and the whole area for loose specimens is tracked along
both the dilatant proportional strain loading path and the biaxial loading path.
The results as function of the mean stress p are presented in Fig.3.11 - 3.14. Small tri-
angles and squares highlight the start and the end points of the curves and a fitting curve
is given based on proportional strain tests in each figure. Two significant characteristics
manifest themselves again: (1) the measures of ri∗ (i ∈ [3, 4, 5, 6p]) from proportional strain
tests converge to form a master curve independent of the value of λ; (2) the master curve
agrees very well with the critical states characterized by the grain loop indexes obtained
0.45
0.40
0.35
0.30
0.25
r3*
0.20
0.15
100kPa-dense ( = 1.2)
0.10 100kPa-dense ( = 1.3)
100kPa-dense ( = 1.4)
0.05 CS ( = 1.2)
0.00 0 20 40 60 80 100 120 140 160
p (kPa)
(a)
0.45
0.40
0.35
0.30
0.25
r3*
0.20
0.15 60kpa-dense ( 0 = 60kPa)
0.10 40kpa-loose ( 0 = 40kPa)
0.05 100kpa-loose ( 0 = 100kPa)
0.00 0 25 50 75 100 125 150 175
p (kPa)
(b)
Figure 3.11: p − r3∗ for proportional strain (a) and biaxial (b) test. The start and end
points are marked by triangles and squares, respectively. An exponential fit is shown for the
0.6
100kPa-dense ( = 1.2)
100kPa-dense ( = 1.3)
0.5 100kPa-dense ( = 1.4)
CS ( = 1.2)
0.4
r6p*
0.3
0.2
0.1
0 20 40 60 80 100 120 140 160
p (kPa)
(a)
0.6
0.5 100kpa-dense ( 0 = 100kPa)
0.4 Proportional strain tests fitted line
r6p*
0.3
0.2
0.1
0 25 50 75 100 125 150 175
p (kPa)
(b)
Figure 3.12: p − r6p∗
for proportional strain (a) and biaxial (b) tests. The start and end
0.45
0.40
0.35
0.30
0.25
r4*
0.20
0.15
100kPa-dense ( = 1.2)
0.10 100kPa-dense ( = 1.3)
100kPa-dense ( = 1.4)
0.05 CS ( = 1.2)
0.00 0 20 40 60 80 100 120 140 160
p (kPa)
0.45
0.40
0.35
0.30
0.25
r4*
0.20
0.15 60kpa-dense ( 0 = 60kPa)
0.10 40kpa-loose ( 0 = 40kPa)
0.05 100kpa-loose ( 0 = 100kPa)
0.00 0 25 50 75 100 125 150 175
p (kPa)
dilatant proportional strain tests and repeated on the biaxial tests, as the all proportional
0.250
0.225
0.200
0.175
0.150
r5*
0.125
100kPa-dense ( = 1.2)
0.100 100kPa-dense ( = 1.3)
100kPa-dense ( = 1.4)
0.075 CS ( = 1.2)
0.050 0 20 40 60 80 100 120 140 160
p (kPa)
0.250
0.225
0.200
0.175
0.150
r5*
0.125 40kpa-dense ( 0 = 40kPa)

0.100 40kpa-loose ( 0 = 40kPa)
0.075 100kpa-loose ( 0 = 100kPa)
0.050 0 25 50 75 100 125 150 175
p (kPa)
in biaxial tests. These results are in agreement with the premise of ACST.
For the shortest loop l3 , r3∗ continuously decreases from the initial to the ultimate point
along all different loading paths. As for the most deformable type of loop l6p , it is in a quite
different situation as shown in Fig.3.12. The value of r6p ∗
for the dense samples increases
during loading, but it decreases for the loose samples.
Different to the fabric tensor based on the statistics of inter-particle contact normals,
investigated in Subsection 3.4.1, that has been widely accepted as a quantitative measure
of fabric anisotropy in granular materials (Li and Dafalias 2015), the scalar ri∗ is rather
a measure characterizing the deformability of granular materials, namely the void ratio.
Unlike the fabric anisotropy normalized by the specific volume (Li and Dafalias 2015), the
evolution of ri∗ depends on p as illustrated in Fig.3.11 - 3.14.
3.5 Mixed proportional strain and biaxial loading paths
A complex loading path combining proportional strain and biaxial paths has been simu-
lated to confirm that proportional strain tests can indeed determine the position of critical
state in the e − p − q space as shown in Fig.3.6 and Fig.3.9. It was conducted as follows:
first, the proportional strain path with λ = −1.2 was imposed; six groups of state data at
ε2 = 0.0022, 0.0065, 0.0109, 0.0250 and 0.0294 were selected; then, a biaxial loading path
was performed at each state while the lateral stress was maintained as it was. Such mixed
loading paths allow us to assess the existence of critical states according to the definition
given in the introduction (continuous shearing with no change in volume or mean pressure,
while keeping aligned stress anisotropy with fabric anisotropy). The stress paths along the
mixed paths are shown in Fig.3.15, while the stress and strain responses are presented in
Fig.3.16.
In Fig.3.15, the six transition states are labelled A, B, C, D, E and F and are marked
by zoomed dots. The average critical (p, q) state is marked by squares. It can be observed
that all six squares from continued biaxial tests lie along the stress track of proportional
strain tests. As seen in Fig.3.16 , when the loading path was adjusted to a biaxial test at
ε2 = 0.0022, a small compaction, followed by a large dilatancy, appears before a steady
regime was reached; when the biaxial test was imposed at ε2 = 0.0065 or 0.0109, the vol-
ume increased at first before it reacheed a steady state; when ε2 = 0.0250 or 0.0294, the
following biaxial loading path fluctuated around constant values for both deviatoric stress
3.5. Mixed proportional strain and biaxial loading paths 47
140
Proportional strain loading D
(A) Transition at 2 = 0.0022
120 (B) Transition at 2 = 0.0065 C
(C) Transition at 2 = 0.0109
(D) Transition at 2 = 0.0207 B E
100 (E) Transition at 2 = 0.0250
(F) Transition at 2 = 0.0294 F
80
q (kPa)
A
60
40
20
0 0 20 40 60 80 100 120 140 160

p (kPa)
Figure 3.15: p − q evolution along combining proportional strain and biaxial loading paths.
The proportional strain loading with λ = −1.2 are conducted at first until a certain axial
strain level labeled as A, B, C, D, E and F, after that a biaxial loading is performed by
keeping the corresponding lateral stress unchanged. The transition states and the average
critical states are marked by dots and squares, respectively. The corresponding stress-
strain responses are shown in Fig.3.16
(A) Transition at 2 = 0.0022 (B) Transition at 2 = 0.0065

150 q of proportional strain loading
4 150 4
q of continued biaxial loading 3 3
100 2 100 2
q (kPa)
q (kPa)
(%)
(%)
1 1
v
50 50
v of proportional strain loading0 0
v of continued biaxial loading
0 0 5 10 15 1 0 0 5 10 15 1
(%) 2 2(%)
(C) Transition at 2 = 0.0109 (D) Transition at 2 = 0.0207
150 4 150 4
3 3
100 2 100 2
q (kPa)
q (kPa)
(%)
(%)
1 1
v
v
50 50
0 0
0 0 5 10 15 1 0 0 5 10 15 1
(%) 2 2 (%)
(E) Transition at 2 = 0.0250 (F) Transition at 2 = 0.0294
150 4 150 4
3 3
100 2 100 2
q (kPa)
q (kPa)
(%)
(%)
1 1
v
50 50
0 0
0 0 5 10 15 1 0 0 5 10 15 1
2 (%) 2 (%)
Figure 3.16: Stress-strain responses along combined proportional strain and biaxial loading
paths. The switch states and ultimate states are marked by dots and squares, respectively.
The corresponding stress paths are shown in Fig.3.15
Figure 3.17: The relation between imposed dilatancy from proportional strain test and
unconstrained dilatancy if switched to a biaxial condition according to results in Fig.3.15
and Fig.3.16. Note that soil mechanics convention is adopted with positive compression.
Figure 3.18: The relation between imposed dilatancy/contractancy in proportional strain

tests and unconstrained dilatancy/contractancy expected in biaxial tests in the plane of ax-
ial strain vs. volumetric strain (left) and the possible stress paths in the plane of deviatoric
stress vs. mean stress (right). The major solid lines represent the imposed dilatancy/con-
tractancy, namely the proportional strain loading paths; the fins denote the incremental
unconstrained dilatancy/contractancy if the loading is switched to biaxial conditions.
and volumetric strain, which means that the switching points correspond to a critical state.
The stress-strain evolution of biaxial tests further proves that at relatively large deforma-
tion, the proportional strain loading path will evolve along the critical state line defined in
biaxial tests.
Based on the results in Figures 3.15 and 3.16, the definition of critical state can be ex-
pressed as follows: for a sample with fixed stress and strain rate directions, a p − q − e state
is considered as a critical state if the application of a biaxial loading under the same lateral
pressure, starting from this p − q − e state, leads to zero volume strain and zero deviatoric
stress evolution. Note that for such a state, a rotation of the principal stress direction with
p and q constant will nevertheless result in a change of e. This result observed for instance
in Theocharis et al. 2019 shows that the critical state cannot be described only by the
three internal state variables p − q − e; additional variables related to the microstructural
anisotropy due to loading direction are necessary to define a state equation for the critical
state line (Li and Dafalias 2012).
Figures 3.15 and 3.16 also provide information for comparing the unconstrained dila-
tancy obtained in biaxial tests with the imposed dilatancy in proportional strain tests 2 .
As illustrated in Fig.3.17, the solid line represents the imposed dilatancy with respect to
axial strain. The fixed dilatant rate is characterized by the slope of the line. The fins give
the unconstrained dilatancy rate (namely a direction of incremental volumetric strain),
that would be observed if the loading were switched to a biaxial one. At the beginning,
the imposed dilatancy is larger than the unconstrained dilatancy (from point O to X),
the mean stress p decreases slightly from p0 to X. When the imposed dilatancy is smaller
than the unconstrained dilatancy expressed as ε̇unconstrained
v < ε̇imposed
v < 0 (from point X
to Y), the stress response p increases from X to Y. The unconstrained dilatancy decreases
continuously thanks to the imposed dilatancy. The states at ε2 = 0.0065 and 0.0109 in
Fig.3.16 belong to this domain. At points X and Y, where the unconstrained dilatancy
equals the imposed dilatancy, the mean stress p reaches the peak. The state at ε2 = 0.0207
in the proportional strain test is close to point Y. After point Y, the unconstrained dila-
tancy decreases continuously to become smaller than the imposed dilatancy, described as
0 > ε̇unconstrained
v > ε̇imposed
v till zero unconstrained dilatancy at point Z (ε̇unconstrained
v = 0).
At point Z, the stress path reaches the CSL defined by biaxial tests. If a biaxial path is
followed after point Z, ε̇unconstrained
v = 0 occurs as shown in the results from mixed tests
switched at ε2 = 0.0250 and 0.0294. The further imposed dilatancy after point Y leads
the stress path to turn back to approach point O along the CSL. In conclusion, when the
imposed dilatancy is smaller (larger) than the unconstrained dilatancy, the mean stress p
2
Note that the terms unconstrained and imposed are used to recall that the volumetric strain is a
response variable in a biaxial loading while it is a control variable in proportional strain tests.
increases (decreases).
The results in Fig.3.7 suggest that the relation between imposed dilatancy/contrac-
tancy in proportional strain tests and dilatancy/contractancy expected in biaxial tests can
be categorized as shown in Fig.3.18. The strain path 1 represents a situation for which
the imposed dilatancy is larger than the unconstrained dilatancy (contractancy can be
regarded as the negative dilatancy) from the beginning. This strain path leads to a lique-
faction presented as stress path 1. Dilatant proportional strain loading path 2 leads to a
more complex response, as discussed before.
The proportional strain test with zero volume change (often referred to as undrained
test) can be categorized as one of two types, depending on whether ε̇unconstrained
v < 0 at the
beginning (path 3) or ε̇v
unconstrained
> 0 (path 4). The unconstrained dilatancy along load-
ing path 3 tends toward zero (horizontal direction), when approaching the critical state.
Conversely, loading path 4, representing an isochoric test on a loose sample, results in
liquefaction.
As for the contracting proportional strain path, two groups can be identified according
to whether the stress path crosses the CSL as paths 5 and 6 do. At the beginning of curve
5, the unconstrained contractancy being larger than the imposed contractancy, the mean
stress p decreases slightly. Shortly afterwards, the unconstrained contractancy becomes
smaller than the imposed contractancy, which leads to the increase in p. The direction
of the fins tending to become horizontal is synchronized with the p − q curve approach-
ing the CSL when strain localization takes place within the sample. The representation
of curve 5 is based on results from simulations of mixed biaxial/contracting proportional
strain loading paths (these results are not presented in this chapter whose focus is upon
dilatant proportional strain loading paths). Curve 6 represents the situation where the
unconstrained contractancy is always smaller than the imposed contractancy, which leads
to a continuous increase in p.
To summarize, the mean stress will increase in the dilatancy domain, when the imposed
dilatancy rate is smaller than the unconstrained dilatancy rate. When the imposed dila-
tancy is larger than the unconstrained dilatancy, the mean stress decreases and tends to
0 for large volumetric strain. When the imposed dilatancy is first smaller and then larger
than the unconstrained dilatancy, the stress path is likely to undergo a stress loop. In
the contractancy domain, when the imposed contractancy is larger than the unconstrained
contractancy, the mean stress increases. On the other hand, an imposed contractancy be-
ing smaller than the unconstrained contractancy will result in a decrease in p.
These numerical findings can also be interpreted in terms of dilatancy angles if we recall
that the dilatancy angle is defined in 2D by sinψ = ε̇v /ε̇d = ε̇v /(2ε̇axial − ε̇v ). As long as
it can be assumed that granular materials follow a standard non-associated elasto-plastic
behavior (which is true in 2D or axisymmetric conditions for instance Nicot and Darve
2007), the dilatancy angle characterize the flow rule, and the unconstrained dilatancy can
be viewed as a material property. This broadens the scope of the analysis summarized
in Figure 3.18 insofar, as making it independent of the loading paths considered (at least
for 2D or axisymmetric conditions). The present results are also consistent with previous
experimental and theoretical studies (Daouadji et al. 2017; Darve 1996). In (Daouadji
et al. 2017), the dilatancy rate defined by ε̇v /ε̇axial can be related to the effective mean
pressure. In the criterion of liquefaction given in Darve 1996, the dilatancy angle obtained
from the flow rule and the mean pressure at a given initial void ratio have a non-linear
negative correlation.
3.6 Conclusion and outlook
By simulating proportional strain tests, biaxial tests and complex loading paths combining
biaxial and proportional strain paths using DEM, we have explored through a series of
mechanical and fabric indexes the relation between proportional strain tests and critical
states from biaxial tests.
In considering the results obtained from biaxial tests, we have observed that all struc-
tural variables (e∗ , αc∗ , ri∗ ) within shear bands for specimens experiencing a localized kine-
matic pattern will converge to the same values as for specimens experiencing a diffuse
kinematic pattern at critical state along the same biaxial loading path. These results are
consistent with the conclusion drawn by Zhu et al. 2016 that localization and diffuse pat-
terns share the same fabric properties.
As for proportional strain tests, when the dilatant rate is relatively small and the axial
strain is large enough, a stress path in the p − q plane will finally reach the critical state
line defined by biaxial tests after experiencing a stress loop. The curves on the p − e∗ (void
ratio), p − αc∗ (fabric anisotropy intensity defined from contact normal direction) and p − ri∗
(population of grain loops) planes also converge towards a master curve regardless of the
magnitude of dilatancy characterized by λ = −1.2, −1.3, −1.4. In combining these results,
it can be inferred that any dilatant volume change will always result in similar (p, q, e∗ ,
3.6. Conclusion and outlook 53
αc∗ , ri∗ ) states after the material memory has been erased.
More interestingly is the relation observed between the mechanical states reached along
a proportional strain test and the critical states obtained from biaxial tests within the
framework of ACST. Macroscopic responses (p, q, e∗ ) and fabric-related indices (αc∗ , ri∗ ) in
homogeneous domains along biaxial tests evolve towards the evolution curve of those vari-
ables from proportional strain tests at critical state. Given these results, only one single
test is therefore necessary to construct the classical critical state line in (p - q - e∗ - αc∗ - ri∗ )
space for any granular material. Therefore, the CS concept can be generalized to a wide
class of loading paths which shows that CS acts as a general attractor irrespective of the
loading path considered. This idea of proportional strain paths enriched considerably the
data base for confirmation of CST and ACST.
The mixed proportional strain and biaxial loading paths also confirm this relation. A
more general definition of CS can be given as follows: a p − q − e state, for a sample with
constant stress and strain rate directions, after a given loading history, is considered at
critical state if the application of a biaxial loading under the same lateral pressure start-
ing from this p − q − e state leads to zero volume strain and zero deviatoric stress evolution.
In addition, the comparison between the imposed dilatancy/contractancy along propor-

tional strain paths and unconstrained dilatancy/contractancy expected along biaxial paths
provides valuable information to the interpretation of the various loading paths obtained
in p − q plane for proportional strain loading paths. Along proportional strain tests, the
volumetric strain, which tends to be steady, produces the stress path approaching the CSL.
Proportional strain loading paths are attracted by the critical state line. For samples
initially at critical state (after a biaxial loading for instance), a proportional strain loading
imposes the evolution of the sample state along the critical state line. Even for samples not
initially at critical state, when subjected to proportional strain loading, the microstruc-
ture rearranges itself and reaches geometrical arrangements corresponding to critical state.
This memory fading is probably a key ingredient affecting complex systems and may also
be at the origin of emerging properties. It will be of great interest to further investigate
the underpinning mechanisms of memory fading according to the kinematic pattern (lo-
calization vs. diffuse mode). In addition, from a microscopic point of view, critical state
is a dynamic equilibrium, where only statistics are constant. Most constitutive models for
granular materials describe the critical state from a phenomenological point of view ignor-
ing the micromechanical dynamics. Allowing critical state to emerge in micromechanical
constitutive models from a physical viewpoint will provide a challenging topic for the fu-
ture. One limitation of this study is that all conclusions above are drawn based on 2D
simulations. The extension to 3D, numerically and experimentally, will be attempted in
the future.
Chapter 4
Dynamical view of the critical state
Contents
4.1 Numerical setup . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 58
4.2 The so-called critical state . . . . . . . . . . . . . . . . . . . . . . . 58
4.2.1 Stress-strain analysis . . . . . . . . . . . . . . . . . . . . . . . . . . . 58
4.2.2 Force chains and deviatoric stress . . . . . . . . . . . . . . . . . . . . 59
4.2.3 Grain loops and volumetric strain . . . . . . . . . . . . . . . . . . . . 61
4.3 Hidden dynamics at critical state . . . . . . . . . . . . . . . . . . 63
4.3.1 Generating and vanishing process of chained particles . . . . . . . . 63
4.3.2 Generating and vanishing process of grain loops . . . . . . . . . . . . 69
4.4 Microstructure reorganization dynamics under different confin-
ing pressures . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 69
4.5 Microstructure dynamics under evolving conditions: memory ef-
fects in granular materials . . . . . . . . . . . . . . . . . . . . . . . 81
4.5.1 Fading process of the initial memory . . . . . . . . . . . . . . . . . . 82
4.5.2 Memory fading process along proportional strain test . . . . . . . . . 87
4.6 Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 93
Stationary regimes exist in a great variety of complex systems, which comprise large
numbers of interacting units (Gardner and Ashby 1970; May 1972; Papadopoulos et al.
2018). At the macroscopic scale, such systems may be characterized by statistical descrip-
tors that remain constant over time, while underlying local scale interactions between com-
ponents continually change. In the case of granular materials, where the internal degrees
of freedom are large, a multitude of conformations (arrangements of particle structures)
and conformational transitions (Tordesillas et al. 2012; Walker et al. 2015; Milo et al.
2002; Tordesillas 2007; Matsushima and Blumenfeld 2014; Walker et al. 2014; Behringer
and Chakraborty 2018; Liu et al. 2020a; Pucilowski and Tordesillas 2020) exist for which
the macroscopic stress and porosity remain constant under continuous shearing. As em-
phasized in the previous chapters, such macroscopic states, named critical states (CS) in
55
56 Chapter 4. Dynamical view of the critical state
the geomechanics community, form the focus of Critical State Theory (CST) (Casagrande
1936; Roscoe, Schofield, and Wroth 1958; Been, Jefferies, and Hachey 1991) which stands
as a cornerstone of constitutive theory for granular materials. Constitutive laws built on
CST are however phenomenological and cannot account for the underlying conformational
transitions which are responsible for the emergence of critical states. To this respect, only
multiscale models may capture the microscale detailed balance responsible for critical state.
But, many aspects of these transitions remain poorly understood, and no multiscale model
exists in which i) the minimal set of conformational transitions responsible for critical state
and ii) the rate of the reorganisation are embedded. Indeed, novel structures, transitions
and regimes are still being uncovered in granular materials under shear, mostly thanks to
the use of Discrete Element modeling (DEM), as well as advanced experimental equipment
(Rocks, Liu, and Katifori 2021).
One line of advance, which is pursued in this chapter, is to adopt a dynamical sys-
tems theory (DST) approach to improve fundamental understanding of the critical state
by focusing on conformational transitions from the perspective of two important classes
of mesostructures, namely force chains and grain loops (Tordesillas et al. 2012; Walker
et al. 2015; Pucilowski and Tordesillas 2020). Here we similarly employ a DST approach
to uncover insights on these structures’ lifespan dynamics. Our approach can help enrich
multiscale constitutive models, such as the H-model (Nicot and Darve 2011b) which can
incorporate the physics of grain loops and force chains to the extent that the emergent
critical states can be reproduced. More broadly, a proper understanding of the birth-death
dynamics of these structures benefits a gamut of applications where granular materials are
subjected to shear, including: the reduction of energy consumption and greenhouse gas
emissions in processes like comminution (Ben-Nun, Einav, and Tordesillas 2010; Walker
et al. 2011), discovery and design of granular materials with enhanced properties, and geo-
hazard forecasting (Tordesillas et al. 2021), to name a few examples.
Prior studies have shown that the emergent mesostructures of force chains and grain
loops are important building blocks for self-organization in 2D and 3D systems (Drescher
and De Jong 1972; Majmudar et al. 2007; Antony 2007; Tordesillas, Walker, and Lin 2010;
Walker et al. 2015; Pucilowski and Tordesillas 2020). Prior work has highlighted that some
statistical descriptors of the microstructure (e.g. fabric tensor, proportions of chained par-
ticles, proportions of the different categories of loops) reach constant values at critical state
(Rothenburg and Kruyt 2004; Fu and Dafalias 2011; Kruyt and Rothenburg 2014; Zhu et
al. 2016; Wang et al. 2020; Deng et al. 2021c; Tordesillas, Walker, and Lin 2010). Indeed
birth-death dynamics and related concepts like “lifespan”, is a promising way to think of
the critical state regime. Granular shear simulations and experiments, including those fo-
cusing on stick-slip dynamics, have shown that this regime is governed by the competing
57
mechanisms of growth/birth versus collapse/death of force chains and these events drive
evolution towards, and during, stationary states of systems under shear (Rechenmacher,
Abedi, and Chupin 2010; Tordesillas et al. 2012; Kuhn 2016b; Kuhn 2016a; Pucilowski and
Tordesillas 2020). Experimental evidence of detailed balance in sheared dense planar gran-
ular systems is recently reported in Sun et al. 2021. On the other hand, a DST analysis of
DEM simulations for a 3D system in the critical state regime uncovered bistable dynamics
(Pucilowski and Tordesillas 2020), which is governed by imperfect detailed balance and
underlying conformational transitions having connections to the rattler dynamics reported
in Wautier, Bonelli, and Nicot 2019b. Despite the significant attention paid to the topic,
the open question(s) that arise from all these studies remains: Why do granular systems
evolve towards a stationary limiting state under continued shear? What mechanisms at-
tract the system towards the critical state?
To quantify whether jamming dominates unjamming in evolving systems, the biology-

inspired concepts of lifespan and life expectancy are particularly useful. Starting from a
reference microstructure conformation, the lifespan of a mesostructure corresponds to the
“time” it has continuously existed since its formation. Its life expectancy corresponds to
the remaining time it will survive before disappearing. By comparing the distribution of
lifespans and life expectancies, one can assess whether the birth rate of a particular class
of mesostructure dominates its death rate. These concepts have been applied to granu-
lar materials for instance by Zhu et al. 2016; Wautier, Bonelli, and Nicot 2018a; Deng
et al. 2021a. Indeed, these concepts enable the quantification of the renewal rate of the
microstructure in both 2D and 3D conditions under an external forcing time parameter.
Specifically, it measures how fast the system is losing the memory of its past state under
an external forcing. Consequently, they are particularly well adapted to treat granular
materials from the perspective of dynamical systems with memory. In all these contexts,
the time may correspond to physical time or any relevant forcing parameter, depending on
the system of interest (for biaxial tests considered in the present study, the axial strain is
the time parameter).
The chapter is organized as follows. In section 4.1, the mechanical responses of granular
specimens under biaxial tests with discrete element modeling (DEM) are analyzed. This
is followed in Section 4.2 by a discussion on critical state from a stationarity perspective.
To this respect, the concepts of lifespan and life expectancy of mesostructures are applied
to force chains and grain loops. Birth and death rates of chained particles and grain
loops are analyzed in Section 4.3. The relations between microstructure dynamics and
loading conditions are presented in Section 4.4. The memory fading processes of the initial
conformation and along the proportional strain tests in granular materials is analyzed in
Section 4.5.
Table 4.1: Initial void ratios e0 of prepared samples.
Sample 40kPa-loose 100kPa-loose 400kPa-loose

e0 0.267 0.261 0.234
Sample 40kPa-dense 100kPa-dense 400kPa-loose
e0 0.197 0.191 0.164
4.1 Numerical setup
A set of biaxial simulations are adopted. Confining pressures of 40kPa, 100kPa and 400kPa
are considered. Note that stresses are obtained by considering an out of plane arbitrary
dimension equal to 0.04 m. Dense and loose samples are prepared as indicated in Table
4.1. With p0 equal 40 kPa, 100 kPa and 400 kPa, the ratio of the average contact overlap
and particle size < un > /d50 are 6.6 × 10−4 , 1.6 × 10−3 and 5.6 × 10−3 , respectively. For
the loose samples (the friction angle is set to 35◦ ), the values of < un > /d50 are 9.2 × 10−4 ,
2.1 × 10−3 and 7.6 × 10−3 . More details about sample preparation can be found in the
previous chapter.
As the previous chapter, the stress and strain states are described in 2D conditions,
with the deviatoric stress being equal to q = σ2 − σ1 , and the volumetric strain being
defined as εv = ε1 + ε2 , where σ1 and σ2 are the principal stresses and ε1 and ε2 are the
principal strains. Axes 1 and 2 refer to lateral and vertical directions, respectively.
4.2 The so-called critical state
In the literature, it is common to report fluctuations of the deviatoric stress and the
volumetric strain around steady-state values rather than a smoothly constant evolution
under biaxial loading. The evolution of micro and meso indices like fabric tensor and ratio
of loops population also exhibits fluctuations at critical state. In this section, macroscopic
deviatoric stress and volumetric strain are put in relation with the evolution of force chains
and grain loops populations, respectively.
4.2.1 Stress-strain analysis
Macroscopic responses of the tested specimens are illustrated in Fig. 4.1, in which devia-
toric stress and volumetric strain evolve with respect to the axial strain. The figure shows
4.2. The so-called critical state 59
typical stress and strain responses of dense and loose samples under the biaxial loadings.
In the dense cases, the deviatoric stress q increases at first then decreases to a stationary
value, while the specimen contracts first and then dilates before reaching a steady state.
With an increase in the confining pressure there is an increase in the shear strength and
a decrease in the dilation. On the other hand, the deviatoric stress and volumetric strain
in the loose samples evolve monotonously in the course of loading. It can be observed
that, at critical state, both deviatoric stress q and volumetric strain εv fluctuate around a
constant value. Based on these mechanical responses, both the stationarity and dynamical
properties of CS will be investigated in a micro-mechanical framework.
The incremental deviatoric strain distribution can be analyzed to characterize kine-

matic patterns, as introduced in literature (O’Sullivan, D. Bray, and Li 2003; Zhu et al.
2016). The maps are computed based on the interpolation of particle incremental dis-
placements over meshed specimen (Liu, Nicot, and Zhou 2018; Zhu et al. 2016). When
dense specimens are considered, shear bands may develop, traversing the whole specimen
as shown in Fig. 4.2. It is the spatial domain within shear bands for dense cases and the
whole sample for loose cases that are considered in this study (Fu and Dafalias 2011; Zhu
et al. 2016). The spatial extension of the shear band was determined based on the method
proposed in (Liu, Nicot, and Zhou 2018).
4.2.2 Force chains and deviatoric stress
Force chains have attracted considerable interest owing to their significant role in force
transmissions in granular systems (Drescher and De Jong 1972; Radjai et al. 1996). Three
conditions proposed by Peters et al. 2005, as recalled in Chapter 2, are adopted to detect
a force chain .
In this chapter, the particles belonging to force chains are called chained particles. The
proportions of chained particles is given by rcha = Ncha /Ntot , where Ncha and Ntot are
the number of chained particles and the total number of particles within the investigated
domain, respectively. The ratio within the shear band rcha∗
and the whole sample rcha are
compared in Fig. 4.3.
It can be seen that the evolution with strain of both rcha ∗

and rcha tends to reach a
constant value (with fluctuations) when the deviatoric stress and the volumetric strain
become constant at CS. rcha
∗
is larger than rcha after the generation of the shear band, with
80 2 80 2
60 60
q (kPa)
q (kPa)
0 0
(%)
(%)
40 40
v
20 q v 20
00
2 00
2
5 10 15 5 10 15
2 (%) 2 (%)
200 2 200 2
150 150
q (kPa)
q (kPa)
0 0
(%)
(%)
100 100
v
v
50 50
00
2 00
2
5 10 15 5 10 15
(%)
2 (%)
2
800 2 800 2
600 600
q (kPa)
q (kPa)
0 0
(%)
(%)
400 400
v
200 200
00
2 00
2
5 10 15 5 10 15
2 (%) 2 (%)
Figure 4.1: Evolution of deviatoric stress q and volumetric strain εv with respect to the
axial strain ε2 along biaxial tests. Note that soil mechanics convention is adopted with
positive compression and positive contraction.
4.2. The so-called critical state 61
Figure 4.2: Incremental deviatoric strain (IDS) maps at the critical states, estimated for
axial strain increments of 0.1%. Two samples are considered: (a) 100kPa-dense sample
(ε2 = 6.1%) and (b) 100kPa-loose sample (ε2 = 8.02%).
greater fluctuations. This may be results in a stronger rearrangement of chained particles

within the shear band. It is worth noting that the ratio of chained particles fluctuating
around a mean constant value highlights that important force chain reorganizations take
place at the critical state.
4.2.3 Grain loops and volumetric strain
In a 2D granular material, meso-loops are composed of a set of contacting grains forming

a closed loop. It is usually categorized according to the edge number (from L3 with three
particles to L6p for loops containing six or more particles), a factor which has a significant
influence on the behavior of a loop, especially its deformability. The percentage of each
category reaches a steady state that corresponds to the critical state regime at the spec-
imen scale (Zhu et al. 2016). In this section, we link the grain loop populations (L3, L4,
L5, L6p) to the volumetric evolution along the biaxial test by introducing the volumetric
strain of the different grain loops.
The incremental strain tensor of each type of loop i can be expressed as follows:
1 X k
dε̄¯Li = |ΩLi |dε̄¯kLi i ∈ (3, 4, 5, 6p) (4.1)
|Ω| k
where Ω is the volume of the global domain considered and ΩkLi is the volume of a
100kpa-dense
36 180
35 160
34 140
33 120
100
rcha (%)
q (kPa)
32
80
31 60
30 40
rcha
*
29 q rcha 20
28
0.00 0.02 0.04 0.06 0.08 0.10 0.120
(a) 2
100kpa-loose
36 180
35 160
34 140
33 120
100
rcha (%)
q (kPa)
32
80
31 60
30 40
29 q rcha 20
28
0.00 0.02 0.04 0.06 0.08 0.10 0.120
(b) 2
Figure 4.3: Proportions of chained particles.Two samples are considered: (a) 100kPa-dense
sample and (b) 100kPa-loose sample. Note that rcha∗
represents the value when the spatial
domain of shear band are considered, and rcha when the whole sample is considered. The
deviatoric stresses are recalled.
4.3. Hidden dynamics at critical state 63
given loop k belonging to the category Li . The incremental strain dε̄¯kLi of each loop ‘k’
can be computed assuming a uniform deformation within that loop according to the strain
definition proposed by Liu, Nicot, and Zhou 2018.
The contribution of local incremental strains to the global incremental strain can be
expressed by Equation 4.2:
Z
1
< dε̄¯ >= dε̄¯dS = dε̄¯L3 + dε̄¯L4 + dε̄¯L5 + dε̄¯L6p (4.2)
|Ω| Ω
Figure 4.4 presents the volumetric strain of each category of loops and their integration
during the biaxial loading as well as the evolution of the global volumetric strain.
Figure 4.4 shows a good agreement between the macroscopically computed volumetric
strain and the mesoscopically averaged one. Despite reaching a zero volume change on
average, it can be seen that each loop category reaches constant dilative or contractive
rates at critical state. The dilatancy from L3 , L4 and L5 , and the contractancy from L6p
should be related with the inherent property of each loop: the grains belonging to Li (i ≤ 5)
have limited space to move inward to induce contractancy, whereas L6p loops have larger
void space to move inward. As a result, critical state can only be achieved from a collective
process by loops transforming from one category to another.
4.3 Hidden dynamics at critical state
As highlighted in the previous sections, the jamming and unjamming behavior at CS

includes dynamic mechanisms at the mesoscale (Rechenmacher, Abedi, and Chupin 2010;
Tordesillas et al. 2012; Kuhn 2016b; Sun et al. 2021). To this respect, the dynamical
rate are investigated in this section through computing the lifespan and life expectancy of
chained particles (i.e. particles belonging to force chains) and grain loops. The incremental
evolutions of the sets of chained particles and grain loops are checked for small axial strain
increments of δε2 = 0.1%.
4.3.1 Generating and vanishing process of chained particles
All chained particles are tracked from birth, i.e. the first time they become part of a
given force chain, to death, i.e. the first time they no longer belong to any force chain.
This enables to construct probability density functions (PDF) for the lifespan and the
life expectancy of chained particles at any time of the simulation. The lifespan and life
0.020
0.015
0.010
0.005
0.000
v
0.005
0.010 v l3 v l6p
v l4 v all
0.015
v l5 v
0.020
0.000 0.025 0.050 0.075 0.100 0.125 0.150
(a)
0.02
0.01
0.00
0.01
0.02
v
0.03
0.04 v l3 v l6p
v l4 v all
0.05
v l5 v
0.06
0.000 0.025 0.050 0.075 0.100 0.125 0.150
(b)
Figure 4.4: Volumetric strain of different types of loops and the global assembly during
the biaxial tests in 100kPa-dense sample (a) and 100kPa-loose sample (b).
expectancy PDFs are displayed in Fig. 4.5 for two arbitrary microstructure conformations
(M and N) in the critical state regime. Because the sample is known to experience shear
band localization (see Fig. 4.2), the analysis is restricted to particles belonging to the shear
band in Fig. 4.5.
At the reference conformation M (resp. N) in Fig. 4.5 (a), the PDFs for the lifes-
pan and life expectancy demonstrate the generating and vanishing processes of chained
particles within the shear band domain. Few chained particles existing at M or N comes
from the initial conformation, which proves that the initial arrangement of particles has
been erased when the sample reaches CS. The symmetric shape of the lifespan and life ex-
pectancy PDFs shows that the rearrangement of chained particles reaches an equilibrium
at CS, which is expected for a stationary regime. More interestingly the shape of the PDFs
for the reference conformations M and N is almost the same. Force chains are always rear-
ranging at critical state and the rate of renewal of the set of chained particles is constant.
These features can be observed in 100kPa –loose with lower renewal rates from Fig. 4.5 (b).
The PDFs have been averaged to get rid of statistical noise coming from the finite size
of the chained particle population at different strain levels belonging to the critical regime
(ε2 = 0.0610, 0.0708, 0.0806, 0.0904, 0.1002 and 0.1100 for the 100kPa-dense sample and
ε2 = 0.0605, 0.0708, 0.0802, 0.0906, 0.1005 and 0.1106 for the 100kPa-loose sample). Since
force chains have a limited life duration at critical state, the six microstructure conforma-
tions can be considered as statistically independent (not correlated), which allows us to
build a single PDF from the six data sets. The corresponding lifespan and life expectancy
PDFs are presented in Figure 4.6 with respect to incremental evolution of the axial strain
(∆ε2 ). Thus, ∆ε2 at current state is 0. ∆ε2 > 0 and ∆ε2 < 0 correspond to the lifespan
and life expectancy of chained particles, respectively.
The renewal rates in dense and loose samples appear to be different at first glance, which
would contradict the existence of a unique critical state, whatever the initial sample density.
To find the origin of the apparent faster renewal rate within the shear band in pthe dense
sample, it is meaningful to compare the incremental strain magnitude dεm = dε21 + dε22
in the shear band domain for the dense sample, to the incremental strain magnitude in
the whole sample domain for the loose sample. Similar to the incremental strain of each
type of loops computed in Subsection 3.3, the incremental strain within the shear band
can be computed by averaging the incremental strains of loops within the shear band over
the area of the shear band. When averaged over the whole sample domain dεm is the same
in dense and loose samples at critical state, as shown in Fig.4.7. However, because of the
strain localization in the dense sample, a larger dε∗m develops within the shear band. The
ratio between < dε∗m > and < dεm > at CS inside the shear band is around 2.8. In other
words, an axial strain increment of 1% from the boundaries of the sample imposes a strain
500 0.0150
Lifespan
Life expectancy 200 0.0125
400 Reference configuration
0.0100
300 150 0.0075
0.0050
q (kPa)
q
PDF *
200
v
100 0.0025
100 0.0000
50 0.0025
0
M N
v 0.0050
100 0
0.000 0.025 0.050 0.075 0.100 0.125 0.150
2
(a)
150 120 0.000

Lifespan
125 Life expectancy 0.002
Reference configuration 100
100 0.004
80 0.006
75
0.008
q (kPa)
50 60
PDF
0.010
25
40
0.012
0
M N 20 0.014
25
q v 0.016
50
0.00 0.05 0.10 0.15 0
2
(b)
Figure 4.5: Lifespan and life expectancy PDFs (probability density functions) of chained
particles in two samples: (a) 100kPa-dense sample where only the spatial domain within
the shear band is considered and (b) 100kPa-loose sample with the whole domain con-
sidered. For each sample, two reference conformations belonging to the stationary regime
are adopted, labeled as M and N. The deviatoric stress curve (dot-dashed line) and the
volumetric strain (dashed line) curves are recalled.
300
100kPa-loose
250 100kPa-dense*
200
150
PDF
100
50
00.04 0.02 0.00 0.02 0.04
2
Figure 4.6: The averaged PDFs of chained particles with respect to the incremental strain
evolution (∆ε2 ), over six microstructure conformations at different strain levels belonging
to the critical regime (ε2 = 0.0610, 0.0708, 0.0806, 0.0904, 0.1002 and 0.1100 for the
100kPa-dense sample and ε2 = 0.0605, 0.0708, 0.0802, 0.0906, 0.1005 and 0.1106 for the
100kPa-loose sample).
increment of 2.8% in the shear band domain because of strain localization. This means
that the actual forcing parameter (the “time”) inside the shear band is 2.8 time faster than
for the whole sample.
As a result, the axial strain applied at sample scale is not the right forcing parameter
in the dense case as the strain is 2.8 times more intense in the shear band domain. In order
to account for this observation, we need to stretch the curves of 100kPa-dense* by the
ratio 2.8 to make the lifespan and life expectancy PDF comparable between the dense and
loose sample cases. Then, it can be observed in Fig. 4.8 that the PDF in dense and loose
samples are coinciding. Similar results can be observed for different confining pressures,
as shown in Appendix A. It can be concluded that the strain magnitude (observed in
homogeneous domain of the samples) is the major factor affecting the rate of rearrangement
of chained particles. We can also conclude that the uniqueness of the critical state (namely
whatever the initial density, statistical descriptors converge towards the same value in
biaxial tests of similar confining pressure) that holds for statistical descriptors, also holds
for the underlying dynamical processes.
100kPa-dense*
0.010 100kPa-dense
100kPa-loose
0.008
m
0.006
d
0.004
0.002
0.000 0.025 0.050 0.075 0.100 0.125 0.150
Figure 4.7: The strain magnitudes within the shear band domain (100kPa-dense*), whole
dense sample (100kPa-dense) and the whole loose sample (100kPa-loose). The strain mag-
nitude ratio between 100kPa-dense* and 100kPa-dense is around 2.8.
140 100kPa-loose
120 100kPa-dense*
100
80
PDF
60
40
20
00.04 0.02 0.00 0.02 0.04
2
Figure 4.8: PDFs for chained particles from 100kPa-loose sample and from 100kPa-dense
sample stretched by the strain magnitude ratio 2.8.
4.4. Microstructure reorganization dynamics under different confining
pressures 69
4.3.2 Generating and vanishing process of grain loops
To further understand the critical state dynamics, the generating and vanishing of grain
loops are analyzed for loops composed of 3, 4, 5 and 6 or more particles. Lifespan and life
expectancy PDFs for grain loops are computed. Similar observations as reported in the
previous subsections are visible in Fig.4.9 - 4.11: (1) PDFs show symmetrical shapes at CS
which suggests birth and death rates of grain loops are equal at CS; (2) PDFs within the
shear band from 100kPa-dense coincide with those obtained in the loose case after being
stretched by the appropriate ratio of 2.8. These results extends the conclusion drawn by
Zhu et al. 2016, that the whole loose sample and the inner part of the shear band of the
dense sample share the same statistical distribution of mesostructures once the critical
state regime is reached.
Based on Figures 4.5 - 4.11, a novel way for understanding CS from the perspective
of a mesoscale dynamics can be inferred: the critical state results from the generating
and vanishing of meso-clusters until a dynamic equilibrium is reached under continuous
shearing. Even though dynamic rearrangements permanently occur, at any time, the
material admits the same fabric characterized by the fabric tensor, the distribution of loop
orders etc.. The particle rearrangements at critical state can be quantitatively characterized
by the lifespan and life expectancy of force chains and grain loops. These observations still
hold when other confining pressures (40 kPa and 400 kPa) are considered as shown in
Appendix A.
4.4 Microstructure reorganization dynamics under dif-

ferent confining pressures
In this section, the influence of the confining pressure on the dynamics in the three loose
samples is analyzed. As shown in Figs.4.12 and 4.13, the renewal rate decreases with
the increase in the confining pressure. The increase in confining pressure helps the meso-
structures live longer (with respect to the forcing parameter of the system, i.e. the incre-
mental axial strain).
This is consistent with the fact that larger confining pressure leads to larger elastic
energy storage and larger overlap at contacts, which delays contact opening and contact
sliding.
Loop 3 Loop 4
0.015
Lifespan v
400 Life expectancy 400 200
0.010
q (kPa)
q
PDF *
200 200 0.005
v
100
50 0.000
0 0
0
0.000 0.025 0.050 0.075 0.100 0.125 0.150 0.000 0.025 0.050 0.075 0.100 0.125 0.150 0.005
Loop 5 Loop 6p
400 400
PDF *
200 200
0 0
0.000 0.025 0.050 0.075 0.100 0.125 0.150 0.000 0.025 0.050 0.075 0.100 0.125 0.150
2 2
(a)
Loop 3 Loop 4
300 300 120 0.000
Lifespan q
Life expectancy 100
200 200 0.005
q (kPa)
100 100 60
PDF
0.010
v
40
0 0 0.015
v
20
100
0.000 0.025 0.050 0.075 0.100 0.125 0.150 100 0
0.000 0.025 0.050 0.075 0.100 0.125 0.150
Loop 5 Loop 6p
300 300
200 200
100 100
PDF
0 0
100
0.000 0.025 0.050 0.075 0.100 0.125 0.150 100
0.000 0.025 0.050 0.075 0.100 0.125 0.150
2 2
(b)
Figure 4.9: Lifespan and life expectancy PDFs (probability density functions) of grain loops
in two samples with respect tho the axial strain ε2 : (a) 100kPa-dense sample where only
the spatial domain within the shear band is considered and (b) 100kPa-loose sample with
the whole domain considered. For each sample, two reference conformations belonging to
the stationary regime are adopted. The deviatoric stress curve (dot-dashed line) and the
volumetric strain (dashed line) curves are recalled in the subfigure of Loop 4.
pressures 71
300 Loop 3 300 Loop 4

100kPa-loose
250 100kPa-dense* 250
200 200
150 150
PDF
100 100
50 50
00.04 0.02 0.00 0.02 0.04 00.04 0.02 0.00 0.02 0.04
300 Loop 5 300 Loop 6p
250 250
200 200
150 150
PDF
100 100
50 50
00.04 0.02 0.00 0.02 0.04 00.04 0.02 0.00 0.02 0.04
2 2
Figure 4.10: The averaged PDFs of grain loops with respect to the incremental strain
evolution (∆ε2 ), over six microstructure conformations at different strain levels belonging
to the critical regime (ε2 = 0.0610, 0.0708, 0.0806, 0.0904, 0.1002 and 0.1100 for the
100kPa-dense sample and ε2 = 0.0605, 0.0708, 0.0802, 0.0906, 0.1005 and 0.1106 for the
100kPa-loose sample).
Figure 4.11: PDFs for grain loops from 100kPa-loose sample and from 100kPa-dense sample
stretched by the strain magnitude ratio 2.8.
pressures 73
150 40kPa-loose
100kPa-loose
125 400kPa-loose
100
PDF
75
50
25
00.04 0.02 0.00 0.02 0.04
2
Figure 4.12: Lifespan and life expectancy PDFs for chained particles under different con-
fining pressures with samples 40kPa-loose, 100kPa-loose and 400kPa-loose considered.
Loop 3 Loop 4
200 40kPa-loose 200
100kPa-loose
150 400kPa-loose 150
PDF
100 100
50 50
00.04 0.02 0.00 0.02 0.04 00.04 0.02 0.00 0.02 0.04
Loop 5 Loop 6p
200 200
150 150
PDF
100 100
50 50
00.04 0.02 0.00 0.02 0.04 00.04 0.02 0.00 0.02 0.04
2 2
Figure 4.13: Lifespan and life expectancy PDFs for grain loops with respect to the incre-
mental strain evolution (∆ε2 ) under different confining pressures with samples 40kPa-loose,
100kPa-loose and 400kPa-loose considered.
pressures 75
In order to quantify the rate of microstructure reorganization with respect to the axial
strain, it is interesting to propose a fit for the PDFs presented so far. As a first simple
guess, an exponential fit with one characteristic strain may been proposed as P DF (∆ε2 ) =
e |∆ε2 |/εc . εc accounts for the typical size of the axial strain increment ∆ε2 needed to
a −
εc
renew the whole mesostructures. The smaller εc , the faster the microstructure renewal
under the external forcing ∆ε2 . However, such a fit cannot capture well the shape of the
whole PDFs. Depending on εc , the fit is able to account either for the beginning or for
the tail of the PDFs. This observation motivated the use of a double exponential fit in the
following form:
1 − |∆ε2| 1 − |∆ε2 |
P DF (∆ε2 ) = Pc e εc1 + (1 − Pc ) e εc2 (4.3)
εc1 εc2
where εc1 and εc2 (εc1 < εc2 ) correspond to two characteristic strains accounting for the
dynamic evolution of the system. εc1 and εc2 account for the typical amplitude of the axial
strain increments ∆ε2 needed to renew the mesostructures on the short term and long term,
respectively. The characteristic probability Pc ∈ [0, 1] can show the relative contribution
of the short and long term mechanisms. When Pc < 0.5, it is the long term mechanism
that dominates; the short term mechanism will be more important when Pc > 0.5; the two
mechanisms are evenly matched when Pc = 0.5.
Figs. 4.14 and 4.15 show that double exponential fit is able to account for the shape of
all the lifespan and life expectancy PDFs. The values of εc1 and εc2 for different confining
pressures are presented in Table 4.2 and shown in Fig. 4.16 with respect to the critical
mean stress.
As shown in Table 4.2 and Fig. 4.16, large loops (L5 and L6p) live shorter (faster
reorganization) than small loops (L3 and L4) and force chains. It is consistent with the
assumption that loops connected to force chains open prior to force chain bending (Wau-
tier, Bonelli, and Nicot 2018b).
When it comes to the effect of confining pressures, the renewal rate of meso-structures
which can be characterized by ε1c1 and ε1c2 decreases with the increase in the confining
pressure. This effect differs according to the type of meso-structures. L3 and L4 seem
to be more sensitive to the confining pressures, especially in the long term mechanism,
which could be due to the fact that these small, less compressible loops can store larger
(stabilizing) elastic energy before breaking. On the other hand, large loops (L5 and L6p)
are more compressible, then less sensitive to confining pressure, as the increase in p does
not have as much stabilization effect as for smaller loops.
Figure 4.14: Double-exponential fits given for the chained particle PDFs with respect to
the absolute value of incremental evolution |∆ε2 | under different confining pressures.
Table 4.2: The fitting parameters for different confining pressures.
Sample Parameters Chained particles L3 L4 L5 L6p

40kPa-loose εc1 0.25% 0.30% 0.33% 0.30% 0.25%
εc2 2.29% 1.61% 1.92% 1.81% 1.59%
Pc 0.165 0.229 0.228 0.288 0.290
εc1 /εc2 9.02 5.44 5.78 5.97 6.41
100kPa-loose εc1 0.35% 0.43% 0.41% 0.34% 0.34%
εc2 3.30% 2.57% 2.79% 2.40% 2.25%
Pc 0.146 0.205 0.162 0.209 0.271
εc1 /εc2 9.35 5.92 6.89 7.03 6.71
400kPa-loose εc1 0.61% 0.77% 0.64% 0.59% 0.50%
εc2 5.33% 5.56% 5.60% 4.35% 3.73%
Pc 0.129 0.124 0.113 0.169 0.209
εc1 /εc2 8.76 7.25 8.74 7.41 7.52
pressures 77
Figure 4.15: Double-exponential fits given for the loop PDFs with respect to the absolute
value of incremental evolution |∆ε2 | under different confining pressures.
Figure 4.16: Two characteristic strains (εc1 and εc2 ), the characteristic probability (Pc ) and
the ratio εεc1
c1
with respect to the mean stress at the critical state. The critical mean stress
under different confining pressures (40 kPa, 100 kPa and 400 kPa) are 59 kPa, 143 kPa
and 559 kPa, respectively.
pressures 79
As for chained particles, even though force chains can share particles with large loops,
their stability (and thus lifespan) is strongly dependent on the existence of small order
loops around them. Thus, the effect of confining pressure on force chains lifespan stem
from the different responses of loops under the confining pressure according to their cate-
gory (small vs. large ones), which may explain why the slope of the εc1 and εc2 for chained
particles is smaller than the one for small loops but larger than the one for large loops.
In addition, the long term mechanism increasingly has dominate effect, since Pc , always
being less than 0.5, decreases with the growth of critical pressures.
The form of the proposed fit may appear arbitrary at first glance; but a double exponen-
tial PDF is the signature of the existence of two mechanisms responsible for microstructure
reorganization acting over two different “time” scales. A first mechanism provokes mi-
crostructure reorganizations over small axial strain scales of typical magnitude εc1 , while
a second mechanism induces microstructure reorganizations over much larger axial strain
scales of typical magnitude εc2 .
The existence of two reorganization mechanisms is indeed consistent with the visual-
ization of the incremental deviatoric strain map in Fig. 4.2 for the loose sample. The
map is far from being homogeneous over the whole sample domain, and zones with large
increments of deviatoric strain are visible (in the form of red zones in Fig. 4.2.). Contrary
to shear bands, these zones are not persisting when the axial strain increases. Indeed,
these zones may be seen as shear transformation zones (STZ) (Schall, Weitz, and Spaepen
2007) and characterized by the recent concept of shear chain (Darve et al. 2020).
As a result, we conjecture that εc1 relates to the axial strain needed for a shear transi-
tion zone to develop, while εc2 > εc1 corresponds to the axial strain needed to develop shear
transition zones everywhere in the sample domain. This conjecture can be initially verified
by the color maps of lifespan and life expectancy of loops in the sample 100kPa-loose, as
shown in Figs. 4.17 and 4.18, respectively.
The short and long threshold are set to 4δε2 and 30δε2 , respectively. In Fig.4.17 (a), red
loops are newly-generated at the current conformation. The loops that have lived longer
than 4δε2 are filled in grey. Fig.4.17 (a) shows that the microstructure is experiencing a
series of localized reorganizations (visible as clusters of loops of the same colors) affecting a
limited proportion of the sample domain. In the long term, these series of reorganizations
1.4 > =5 1.4

> 30
1.2 1.2
4
1.0 1.0
3
Lifespan (* )
0.8 0.8 5 30
0.6 2 0.6
0.4 1 0.4
0 4
0.2 0.2
0
0.0 0.0
0.0 0.2 0.4 0.6 0.8 1.0 0.0 0.2 0.4 0.6 0.8 1.0
(a) (b)
Figure 4.17: Color maps of the lifespan of grains loops for the reference axial strain ε2 =
8.02%. Two thresholds are considered corresponding to the two characteristic strains: (a)
0.4% and (b) 3%.
1.4 > =5 1.4

> 30
1.2 1.2
4
1.0 1.0
Life expectancy (* )
3
0.8 0.8 5 30
0.6 2 0.6
0.4 1 0.4
0 4
0.2 0.2
0
0.0 0.0
0.0 0.2 0.4 0.6 0.8 1.0 0.0 0.2 0.4 0.6 0.8 1.0
(a) (b)
Figure 4.18: Color maps of the life expectancy of grains loops for the reference axial strain
ε2 = 8.02%. Two thresholds are considered corresponding to the two characteristic strains:
(a) 0.4% and (b) 3%.
4.5. Microstructure dynamics under evolving conditions: memory effects in
granular materials 81
will affect the whole sample domain as shown in Fig.4.17 (b), where loops that have been
existing not less than 30δε2 are very limited. In Fig.4.18 (a), grey loops are those that will
live longer than four steps. It can also be observed in Fig.4.18 (b) that beyond the long
threshold, only a small part of the loops will still be alive.
4.5 Microstructure dynamics under evolving conditions:

memory effects in granular materials
The critical state has been paid great attention at both macroscopic and microscopic scales
with the emphasis of the stationary section. However, how the CS emerges from the initial
state under shearing from a micromechanical perspective has been highlighted much less
often. Tordesillas 2007 pictured the whole fabric evolution process through introducing
the jamming and unjamming concepts in granular materials under shearing. Shear band
generating and expanding processes are characterized by microstructure dynamics in 2D
simulations. It has been widely accepted that the critical state is independent on the
initial void ratio. Matsushima and Blumenfeld 2017 found that the microstructural char-
acteristics are governed mainly by the packing procedure, and the effects of intergranular
friction and initial states are details that can be scaled away. But the fading process of
the initial memory of the sample is unclear. It has been proved that the sample under the
proportional strain test can evolve along the critical state line normally defined by a series
of biaxial tests in Chapter 3. But how the memory of the structure has been erased needs
further exploration.
It is the external loading and internal rearrangements that work together to form the
stationary state. As commented in Matsushima and Blumenfeld 2017, the self-organization
leaves a fingerprint that manifests in the emergence of intriguing universality-like behav-
ior. The structural characteristics are not completely random and bear the hallmark of
the self-organization process that brought the medium to mechanical equilibrium. Thus,
the purpose of this chapter is to explore the competition between microstructure dynamics
and loading condition, through memory effects characterized by the life duration of grain
clusters. We focus on grain loops and force chains.
Particularly, the life duration of the mesostructures introduced in the previous chapter
is recalled in this section. Three tests are considered: biaxial tests in a dense and a loose
sample, and a dilatant proportional strain test in a dense sample.
4.5.1 Fading process of the initial memory
4.5.1.1 Chained particles
In Fig.4.19, lifespan and life expectancy PDFs for chained particles are displayed for the
initial state of the whole sample in both dense and loose cases.
In Fig. 4.19 (a), it can be observed that a part of the chained particles at the initial
state is transferred to be non-chained particles at an early state (before stress stabilizes)
during the loading, while the other remains continuously in force chains until the end of
the test. The permanent particles will be proved to be located outside of the shear band
domain later in this section.
In general, a chained particle being converted to be non-chained can be caused by (1)

the force chain being transferred to a normal chain without collapse, buckling or bending
because of redistribution of contact forces, or (2) collapse, buckling or bending of the force
chain. It is important to distinguish the two modes. The former one does not relate to
particle rearrangements, while the latter does.
In Fig. 4.19 (a), the decreasing process of the PDF can be divided into three stages
according to the curve slope, namely the rate of transition from chained particles to non-
chained particles: a rapid drop at stage I, a gentle decrease at stage II and nearly no change
at stage III. The transition point between the first two slopes is close to the characteristic
point, which corresponds to the equality between incremental elastic strain energy and
incremental plastic strain energy (Zhu et al. 2016), and the appearance of shear band (Liu
et al. 2020a). The first drop should be mostly related to the stress anisotropy resulting
from the redistribution of contact forces. A number of vertical force chains are built in the
assembly. The second decrease corresponds to the softening process resulting from strain
localization, during which the buckling and bending of force chains occurring within the
shear band, erasing the memory at the initial state. The stationary stage corresponds to
the CS characterized by constant deviatoric stress and volumetric strain. It demonstrates
that once the sample reaches the CS, chained particles from initial state nearly stop disap-
pearing (the remaining ones belong indeed to the sample domain out of the shear band).
As discussed in the previous chapter, few chained particles existing at the critical state
come from the initial state, which demonstrates that the initial arrangement of particles
has been erased when the sample reaches CS within the shear band. Thus, the chained
particles living throughout the whole loading processes almost all belong to the outside
shear band domain.
(a)
300 0.000
Life expactancy qv
160
250 0.002
140
0.004
200 120
0.006
150 100
q (kPa)
PDF
0.008
v
80
100
60 0.010
50 40 0.012
0 20 0.014
0
0.000 0.025 0.050 0.075 0.100 0.125 0.150 0.175
2
(b)
Figure 4.19: Life expectancy PDFs (probability density functions) of chained particles at
the initial state of the whole sample in the dense (a) and loose (b) cases. The deviatoric
stress curve (dot-dashed line) and the volumetric strain (dashed line) curves are recalled.
The decrease in the PDF, namely the vanishing of chained particles at the initial state,
is related to the building up of the stress anisotropy. The assembly builds a vertical force
network. This network will not live long as the lateral dilatation of the sample imposed
by the biaxial loading increasingly open the lateral contacts supporting the force chains.
At the characteristic point, plasticity mechanisms become dominant, force chain bend and
collapse, which results in the gentle decreasing part of the PDF.
By contrast, in Fig.4.19 (b) few chained particles survive from the initial state re-
maining to the end of test in a loose sample. This is consistent with the diffuse failure
mode that grain reorganization occurs everywhere. Comparing the beginning stage of the
two PDF curves, the first drop in the dense sample is stronger and quicker than that in
the loose sample. This is because the strain localization drives an intense and fast reorga-
nization within the band (the strain rate is 3.2 times larger as analyzed in subsection 4.3.1).
4.5.1.2 Grain loops
To further understand the critical state dynamics, the life expectancy of grain loops at the
initial states is analyzed in this subsection. The disappearance of a grain loop composing of
a cluster of grains is confirmed by checking whether some grains come in or out of it in the
next simulation time step. Thus, the life duration of grain loops can be a direct evidence
of structural self-organization. The PDF for life expectancy of grain loops at initial state
is presented in Fig.4.20.
Similar interesting observations that have been reported in the previous Subsection
can be found in Fig.4.20: (1) in the dense sample, around 80% (accumulation of PDF) of
the loops existing at the initial state vanish gradually before the end of the test, and the
rest remains until the end, located outside the shear band; while the loose sample totally
reorganize. (2) The vanishing process is much smoother and gentler in the loose sample
than in the dense sample.
In Fig.4.20 (a), the PDF curve for Loop3 is nonlinear before approaching a steady value
around 0. The strongest vanishing point of L3 locates close to the characteristic point of
the volumetric strain. Change from contractancy to dilatancy corresponding to the break-
age of L3. The non-linearity is becoming gentler with the increase in the order of loops.
As for L6p, the largest vanishing occurs at the initial point. In the dense sample, L6p
is the most sensitive type of loops at the contractive stage. In the loose sample, PDFs
for all types of loops are in similar trends. A Slight difference among them can found in
the initial values. L6p has be largest initial value compared to the others. In both dense
Loop 3 Loop 4
300 300 175
Life expectancy 0.0100
150
200 200 0.0075
125
100 0.0050
q (kPa)
q
PDF
100 100 0.0025
v
75
50 0.0000
0 0
v 25 0.0025
100 100 0
0.00 0.02 0.04 0.06 0.08 0.10 0.12 0.14 0.00 0.02 0.04 0.06 0.08 0.10 0.12 0.14
Loop 5 Loop 6p
300 300
200 200
PDF
100 100
0 0
100 100
0.00 0.02 0.04 0.06 0.08 0.10 0.12 0.14 0.00 0.02 0.04 0.06 0.08 0.10 0.12 0.14
2 2
(a)
Loop 3 Loop 4
300 300 175 0.0000
Life expectancy 150 0.0025
200 200 125 0.0050
100
q (kPa)
q 0.0075
PDF
100 100
v
75
0.0100
0 0 50
0.0125
v 25
100 100 0 0.0150
0.000 0.025 0.050 0.075 0.100 0.125 0.150 0.175 0.000 0.025 0.050 0.075 0.100 0.125 0.150 0.175
Loop 5 Loop 6p
300 300
200 200
PDF
100 100
0 0
100 100
0.000 0.025 0.050 0.075 0.100 0.125 0.150 0.175 0.000 0.025 0.050 0.075 0.100 0.125 0.150 0.175
2 2
(b)
Figure 4.20: Life expectancy PDFs (probability density functions) of grain loops at the
initial state of the whole sample in the dense (a) and loose (b) cases. The deviatoric stress
curve (dot-dashed line) and the volumetric strain (dashed line) curves are recalled in the
subfigure of Loop 4.
>=10 >=120
1.4 1.4
100
1.2 8 1.2
1.0 1.0 80
6
0.8 0.8
60
*
*
0.6 4 0.6
40
0.4 0.4
2
0.2 0.2 20
0.0 0 0.0 0
0.0 0.2 0.4 0.6 0.8 1.0 0.0 0.2 0.4 0.6 0.8 1.0
(a) (b)
>=10 >=120
1.4 1.4
100
1.2 8 1.2
1.0 1.0 80
6
0.8 0.8
60
*
*
0.6 4 0.6
40
0.4 0.4
2
0.2 0.2 20
0.0 0.0
0 0
0.0 0.2 0.4 0.6 0.8 1.0 0.0 0.2 0.4 0.6 0.8 1.0
(c) (d)
Figure 4.21: Color map of life expectancy of grain loops at the initial state under the
biaxial test in the dense (a,b) and loose (c,d) samples. Two thresholds are used in the
color scale: 10 (a,c) and 120 (b,c) times the reference incremental axial strain δε1 = 0.1%
which corresponds respectively to 1 % and 10 % of axial strain from the initial state in the
biaxial loading
and loose samples, l6p is the weakest category at the beginning stage where the deviatoric
stress level is low.
To demonstrate the memory fading process intuitively, the life expectancy characterized
by incremental axial strains has been shown in the color map in Fig.4.21. Figure 4.21 (a)
and (c) presents the loops destroyed within 10 increments δε2 of axial strain (δε2 = 0.1%)
for the dense and loose samples, respectively. A band can be observed in Fig.4.21 (b). Yel-
low loops corresponds to loops alive until the end of the test. They are located outside the
band, coexisting with loops having short life expectancy. Particle rearrangements occur
early and intensely within the shear band, and over a longer time scale outside the shear
band. The life expectancy of mesostructures is a promising index that can be used to iden-
tify the shear band. The vanishing of loops in the loose sample is relatively homogeneous
in space, as shown in Fig.4.21 (d).
4.5.2 Memory fading process along proportional strain test
It has been proved that one single dilatant proportional strain test can present a critical
state line normally defined by a series of biaxial tests. Lifespan dynamics of cluster confor-
mations would help to refine the mesostructure reorganization along the decrease in mean
stress during a proportional dilatant strain loading, thus, to better understand the critical
state line as an attractor.
In Fig. 4.22, life expectancy PDFs (probability density functions) of different types of
grain loops are presented under the dilatant proportional strain loading considering the
whole sample, with deviatoric stress and volumetric strain curves. The decrease in each
PDF is rapid initially and reach at a stationary state close to zero. There is little grain
loops remaining until the end of the test.
The lifespan and life expectancy PDFs at the deviatoric stress descending stage are
explored within the shear band. The two axial strains, 0.0610 and 0.1094, are considered
as reference states, as shown in Fig.4.23. The generating and vanishing processes for grains
loops appear to be symmetrical, which can be further observed in Fig.4.25 presenting the
absolute value of PDFs with respect to the axial strain evolution (∆ε2 ). In addition, lifes-
pan and life expectancy PDFs are explored with the whole sample considered, as shown
in Figure 4.26. Corresponding fits are given based on the double exponential equation
4.3. It can be seen that there is a narrow gap between the lifespan and life expectancy
Loop 3 Loop 4
300 300 175 0.030
Life expectancy 150
200 200 0.025
125
0.020
100
q (kPa)
q
PDF
100 100 0.015
v
75
0.010
0 0 50
25 0.005
v
100 100 0 0.000
0.00 0.02 0.04 0.06 0.08 0.10 0.12 0.14 0.00 0.02 0.04 0.06 0.08 0.10 0.12 0.14
Loop 5 Loop 6p
300 300
200 200
PDF
100 100
0 0
100 100
0.00 0.02 0.04 0.06 0.08 0.10 0.12 0.14 0.00 0.02 0.04 0.06 0.08 0.10 0.12 0.14
2 2
Figure 4.22: Life expectancy PDFs (probability density functions) of grain loops under the
dilatant proportional strain loading considering the whole sample.
PDFs. The gap could be the origin of the leakage as a imperfect balance. The leakage is
believed to mainly happen around the blurred boundary of the shear band area since the
lifespan and life expectancy PDFs within the shear band domain are symmetric. The life
expectancy curve is above the lifespan curve which may due to the decrease in the mean
stress resulting in slower structural reorganization.
A series of life expectancy color maps are shown in Fig.4.27, with the initial state
(a), the peak of deviatoric stress (b), states belonging to the convergent stage (c-f) being
considered as marked in Fig.4.24.
In the first four maps, the shear band can be observed. Within the shear band, the
color become darker and darker from (a) to (d) corresponding to faster and faster rate of
grain rearrangement, which will be further discussed. Outside the shear band, the domain
is similarly darker and darker. However, the shape of the band has little change from (a) to
(d). The band is no more evident at the state (e) and (f) where the vanishing of grain loops
inside and outside the shear band is almost at the same rate. These maps demonstrate
that the process from localization to liquefaction along the dilatant proportional strain test
does not result from the expanding of the shear band to the whole sample, but is caused
by the unloading process with decrease of stresses and the disappearing of the shear band.
Loop 3 Loop 4
500 500 175 0.030
400
Lifespan 400 150
300 300 125
0.020
100
q (kPa)
q
PDF
200 200 0.015
v
75
100 100 0.010
50
0 0 25 0.005
v
100 100 0 0.000
0.00 0.02 0.04 0.06 0.08 0.10 0.12 0.14 0.00 0.02 0.04 0.06 0.08 0.10 0.12 0.14
Loop 5 Loop 6p
500 500
400 400
300 300
PDF
200 200
100 100
0 0
100 100
0.00 0.02 0.04 0.06 0.08 0.10 0.12 0.14 0.00 0.02 0.04 0.06 0.08 0.10 0.12 0.14
2 2
(a)
Loop 3 Loop 4
500 500 175 0.030
400
Lifespan 400 150
300 300 125
0.020
100
q (kPa)
q
PDF
200 200 0.015
v
75
100 100 0.010
50
0 0 25 0.005
v
100 100 0 0.000
0.00 0.02 0.04 0.06 0.08 0.10 0.12 0.14 0.00 0.02 0.04 0.06 0.08 0.10 0.12 0.14
Loop 5 Loop 6p
500 500
400 400
300 300
PDF
200 200
100 100
0 0
100 100
0.00 0.02 0.04 0.06 0.08 0.10 0.12 0.14 0.00 0.02 0.04 0.06 0.08 0.10 0.12 0.14
2 2
(b)
Figure 4.23: Lifespan and life expectancy PDFs (probability density functions) of grain
loops at the convergent state under the dilatant proportional strain loading considering a
domain within the shear band.
140
120
100
q (kPa)
80
60
40
20
00 2 4 6 8 10 12 14 16
(%)
Figure 4.24: The six states chosen for color maps along the proportional strain tests
Loop 3 Loop 4
600 600
Lifespan
500 Life expectancy 500
400 400
300 300
PDF
200 200
100 100
0 0
0.04 0.02 0.00 0.02 0.04 0.04 0.02 0.00 0.02 0.04
Loop 5 Loop 6p
600 600
500 500
400 400
300 300
PDF
200 200
100 100
0 0
0.04 0.02 0.00 0.02 0.04 0.04 0.02 0.00 0.02 0.04
2 2
Figure 4.25: The absolute value of PDFs for Lifespan and life expectancy of grain loops
under the dilatant proportional strain loading within the domain of the shear band.
Loop 3 Loop 4
1000 1000
Lifespan
Fit of lifespan
800 Life expectancy 800
Fit of life expectancy
600 600
PDF
400 400
200 200
0 0
0.01 0.00 0.01 0.02 0.03 0.04 0.01 0.00 0.01 0.02 0.03 0.04
Loop 5 Loop 6p
1000 1000
800 800
600 600
PDF
400 400
200 200
0 0
0.01 0.00 0.01 0.02 0.03 0.04 0.01 0.00 0.01 0.02 0.03 0.04
2 2
Figure 4.26: The absolute value of PDFs for Lifespan and life expectancy of grain loops
under the dilatant proportional strain loading with the whole sample considered. A fit is
given to the PDFs using the double exponential in Equation 4.3.
>=120 >=120
1.4 1.4
100 100
1.2 1.2
1.0 80 1.0 80
0.8 0.8
60 60
*
0.6 0.6
40 40
0.4 0.4
0.2 20 0.2 20
0.0 0 0.0 0
0.0 0.2 0.4 0.6 0.8 1.0 0.0 0.2 0.4 0.6 0.8 1.0
(a) (b)
>=120 120
1.4 1.4
100 1.2 100
1.2
1.0 80 1.0 80
0.8 0.8
60 60
*
*
0.6 0.6
40 40
0.4 0.4
20 20
0.2 0.2
0 0
0.0 0.2 0.4 0.6 0.8 1.0 0.0 0.2 0.4 0.6 0.8 1.0
(c) (d)
120 120
1.4 1.4
1.2 100 1.2 100
1.0 80 1.0 80
0.8 0.8
60 60
*
0.6 0.6
40 40
0.4 0.4
20 20
0.2 0.2
0 0
0.0 0.2 0.4 0.6 0.8 1.0 0.0 0.2 0.4 0.6 0.8 1.0
(e) (f)
Figure 4.27: Color map of life expectancy of grain loops at different conformations: (a)
the initial state ε2 = 0, (b) the peak of deviatoric stress ε2 = 0.0185, (c-f) belonging
to the convergent stage under the dilatant proportional strain test in the dense sample
ε2 = 0.0610, 0.0810, 0.0910, 0.1094.
4.6. Conclusion 93
4.6 Conclusion
The critical state normally emphasized as a stationary state in geomechanical fields was
revisited and characterized based on a series of analysis on the rearrangement of particles
from a dynamic perspective at the mesoscale. We have shown that the critical state results
from the balance competing between generating and vanishing of meso-clusters. We have
established that force chains and grain loops have a regular, short life duration (0.4-3%
compared to the whole 15% loading) at critical state, corresponding to a rapid fading of
the material memory along its loading history. The critical state not just "forgot" the
initial microstructure conformation but its recent past in the stationary regime in a sample
under continuous shearing. We have shown that the uniqueness of the critical state that
holds for statistical descriptors also holds for the underlying dynamical processes.
By fitting the data, we have shown that force chains and loops of different sizes have
different lifespan and life expectancy which increases with the mean stress level. Larger
loops have a shorter life due to their deformability and force chain life duration is similar
to the life duration of small order loops which is consistent with the known relationship
between force chains and their small order supporting loops. Moreover, the microstructure
reorganization has been proved to rely on two mechanisms that act over very different axial
strain increments. A local mechanism related to sheared mesostructures provoke a rapid
(i.e.for small axial strain increments) reorganization of the microstructure while the char-
acteristic strain related to the nucleation of new sheared mesostructures account for the
microstructure reorganization over larger strain increments. The recent concept of shear
chain (Darve et al. 2020) will probably help precise the elastic and plastic mechanisms re-
sponsible for the observed dynamics of critical state. Complementary investigations deserve
being conducted to analyze the velocity of strain waves in the sample, since the dynamic
load transfer characteristics can be also interpreted within the context of wave propagation
theory (Sadd, Adhikari, and Cardoso 2000). Such a future work may help seeing whether
the unjammed microstructure keeps a partial memory of its previous jammed state.
The new findings presented in this chapter might help to find clues to explicitly incor-
porate CS features in micro-mechanical constitutive models, as well as to limit the number
of different loading paths needed to construct the data base in data driven models because
of limited hysteresis effect. In relation with one particular micromechanical model, it
would be possible to consider the rearrangement of the hexagonal loop meso-structures on
the H-model (Nicot and Darve 2011b; Wautier et al. 2021) to achieve a better description
of the critical state, based on the renewal rate of mesostructures as quantified in this study.
If the present study deals with critical state dynamics in granular material, the concept,
tools and methodology are rather generic and may be applied to a great variety of other
evolving complex systems to reveal their hidden dynamics.
Chapter 5
Critical state and the H-model
Contents
5.1 Review and analysis of the H-model . . . . . . . . . . . . . . . . . 97
5.1.1 The H-model in brief . . . . . . . . . . . . . . . . . . . . . . . . . . . 97
5.1.2 Biaxial test at the material point scale . . . . . . . . . . . . . . . . . 102
5.1.3 Mesoscale inspection of the H-model during biaxial loading . . . . . 105
5.2 Emergence of critical state in the H-model . . . . . . . . . . . . . 114
5.2.1 Method . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 114
5.2.2 Preliminary results . . . . . . . . . . . . . . . . . . . . . . . . . . . . 115
5.3 Conclusions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 121
Based on the extensive investigations on the critical state conducted in the previous
chapters, it was shown that mesostructural transformations under external loading is a cen-
tral ingredient that should be included in multiscale constitutive modeling. In this chapter,
the framework of the H-model is reviewed, and inspected at mesoscale on a biaxial test
in Section 5.1. In Section 5.2, a method is proposed to introduce dynamical reorganiza-
tion between generating and vanishing of meso-clusters through a deactivation/reactivation
procedure. This process is inspired by results obtained in Chapter 4. Based on the up-
dated framework of H-model, the global stress-strain in a biaxial test is simulated, and the
corresponding meso-structural analysis is presented. Additionally, predictions under two
different densities and two different confining pressures are compared separately to assess
the performance of the updated H-model with respect to critical state prediction.
95
96 Chapter 5. Critical state and the H-model
(a) (b)
Figure 5.1: The directional distributions and the coordinates at the REV scale (e1 , e2 )
and the mesoscale (n, t).
(a) (b)
Figure 5.2: The symmetrical description of geometry (a) and forces (b) at the elementary
hexagonal pattern of adjoining particles (H-cell). There are two types of contact c1 and cc ,
the corresponding contact branches are d1 and d2 .
5.1. Review and analysis of the H-model 97
Figure 5.3: General homogenization scheme relating both incremental stress and strain
tensors
5.1 Review and analysis of the H-model
5.1.1 The H-model in brief
The H-model, also known as H-directional model, is a micromechanical-based constitutive

model initially proposed by Nicot and Darve 2011b. In the H-model, a granular assembly
is described by a distribution of mesostructures oriented in different directions, as shown in
Fig.5.1 (a). The coordinate system for each hexagon is presented by (e1 , e2 ) in the global
frame and (n, t) in the local frame, as shown in Fig.5.1 (b). The mesostructure (H-cell)
used in the H-model corresponds to a hexagonal pattern in 2D and a bi-hexagonal one in
3D (Nicot and Darve 2011b; Xiong, Nicot, and Yin 2017; Wautier et al. 2021). We focus
on the 2D H-model in this Chapter. The collective rearrangement in granular materials
is accounted in the H-model through the deformation of the H-cells. The geometry and
forces at the H-cell are symmetric, as presented in Fig.5.2. Because of the symmetry of the
H-cell, there are only two types of contact: c1 and c2 at each H-cell. Behaviors at different
scales (the contact scale, the hexagonal scale and the REV scale) are related based on the
homogenization scheme in Fig.5.3. Thus the H-model accounts for geometric effects in the
macroscopic behavior of granular materials.
5.1.1.1 Strain localization hypothesis
The global incremental strain δεij is localized to each hexagon according to the equations
as follows: (
δl1 = −l1 δεij ni nj
(5.1)
δl2 = −l2 δεij ti tj
These formula corresponds to the relative length variation in direction n and t. Note that
no distortion is taken into account in the model for the sake of preserving the symmetry
of the H-cell.
5.1.1.2 Directional behavior
Each H-cell representing one direction can be as illustrated in Fig.5.2. The state of a H-cell
can be defined by three geometrical parameters: the opening angle α, the inter-granular
distances d1 and d2 . These parameters relate to the cell dimensions l1 and l2 as
(
l1 = d2 + 2d1 cosα
(5.2)
l2 = 2d1 sinα,
The mechanical behavior of each hexagon depends on both the contact law between ad-
joining grains and the mechanical equilibrium with external forces. An elastic-frictional
contact law is adopted. Such a contact law relies on three parameters: a normal stiff-
ness kn , a tangential stiffness kt , and an inter-granular friction angle ϕg . The normal and
tangential contact forces can be expressed in an incremental form as
(
δNi = kn δuin
(5.3)
δTi = min[|Ti + kt uit |, tanϕg (Ni + δNi )]ξ − Ti ,
where ξ is the sign of the quantity Ti + kt uit , uin and uit are the normal and tangential
relative displacements of the grains at contact i. Given the symmetries in the H cell, we
only need to consider the contact forces called N1 , N2 and T1 as shown in Fig.5.2, The
corresponding incremental displacements are obtained by differentiating the expression of
the branch vector joining the grains in contact:
δu1n = −δd1
δu2n = −δd2 (5.4)
δu1t = −δd1 δα,
Mechanical balance of grain 2 reads, along directions n:
N2 = N1 cosα + T1 sinα + G2 = N1 cosα + T1 (sinα + iG2 ) (5.5)

where iG2 = 0 when G2 = 0 in the original version of H model (Nicot and Darve 2011b),
iG2 = 1 when G2 = T1 in the updated version (Wautier et al. 2021). Equation 5.5 differen-
tiates into
cosαδN1 − (iG2 + sinα)δT1 − δN2 + (N1 sinα − T1 cosα)δα = 0 (5.6)
Combining these equations at the H-cell scale, the incremental evolution δl1 and δl2 can
be related to δd1 , δd2 and δα in matrix form as
     
δd1 δl1 2cosα 1 −2d1 sinα
A δd2  = δl2  , A =  2sinα 0 2d1 cosα  (5.7)
δα λ A31 −1 A33
where (
cosα if no sliding occurs
A31 = (5.8)
cosα + ξtanϕg (iG2 + sinα) if sliding occurs
(
[N1 sinα − T1 cosα − kt d1 (iG2 + sinα)]/kn if no sliding occurs
A33 = (5.9)
(N1 sinα − T1 cosα)/kn if sliding occurs
(
0 if no sliding occurs
λ= (5.10)
ξ(1 + sinα)(tanϕg N1 − T1 ) if sliding occurs
Based on the updated variables δd1 , δd2 and δα, the contact evolution (δN1 , δN2 and δT1 )
can be computed according to Equation 5.7.
5.1.1.3 Stress averaging
The macroscopic stress tensor σ at a material point is calculated by a directional averag-

ing of all contact forces acting in a collection of unit H-cells (belonging to a virtual REV
corresponding to this material point). This is done in two steps by first defining a meso
stress σmeso for each H-cell and then averaging all the meso-stresses of the H-cell collection.
Here and below, the meso-stress tensor is defined at the scale of the unit H-cell using
the Love-Weber formula:
1 X O
σmeso = fc lc (5.11)
Vmeso c∈V
meso
where fc is the contact force at contact c, lc the branch vector joining the centers of the two
grains in contact and Vmeso the volume of the unit H-cell. Note that Bagi correction term
accounting for boundary contact forces to the H-cell is not included in the computation
here (Bonelli et al. 2012; Nicot et al. 2013b; Yan and Regueiro 2019). This is part of the
model assumption.
Expressed in the local frame (n, t), equation 5.11 gives with soil mechanics conventions
(positive compression).

meso 2
Vmeso σnn = 4N1 d1 cos α + 4T1 d1 cosαsinα + 2N2 d2

meso
Vmeso σtt = 4N1 d1 sin2 α − 4T1 d1 cosαsinα (5.12)

meso meso
Vmeso σnt = Vmeso σtn =0

Finally, the macroscopic stress tensor for the REV can be determined as an integration
of all meso-stresses of individual unit H-cells, knowing their statistical distribution via the
probability density function ω(θ). Thus, for the entire collection of unit H-cells,
1 π
Z
σ= ω(θ)Vmeso (θ)σmeso (θ)dθ (5.13)
V θ=0
such that Z π Z π
V = ω(θ)Vmeso (θ)dθ and ω(θ)dθ = 1 (5.14)
θ=0 θ=0
If σmeso is expressed as a matrix σmeso

n,t
in the local frame n, t, the stress matrix σ in the
global frame e1 , e2 is readily obtained by conversion from local to global configuration, i.e.
1 π
Z
σ e1 ,e2
= ω(θ)Vmeso (θ)P −1 σmeso
n,t
(θ)P dθ (5.15)
V θ=0
where P is the transformation matrix,

cosθ sinθ
P = (5.16)
−sinθ cosθ
The probability density function ω(θ) describes the statistical distribution of unit H-cells
and accounts for possible anisotropy. It is assumed that the initial distribution of unit
H-cells can be approximated for instance as
1
ω(θ) = [1 + αω cos2(θ − βω )] (5.17)
π
where αω is a parameter ranging from 0 to 1 describing anisotropy of the hexagon cell dis-
tribution, and βω its major principal direction. In practice, the distribution is discretized
in nθ directions.
Rπ
The global volume in Equ.5.15 reads as V = θ=0 ω(θ)Vmeso (θ). Note that the choice of
Vmeso is not unique as long as the strain localization equation 5.1 is updated accordingly
to ensure that δVmeso
Vmeso
= T r(δσ). There are three main options for the volume of H-cell as
shown in Fig.5.4. From the smallest to the largest, they are labelled as Vmeso
1
, Vmeso
2
and
Figure 5.4: Three options of the meso-volume: hexagonal domain (blue), rectangular do-
main going through the grain centers (green), bounding box domain (red)
3
Vmeso . With these different meso-domains being considered, there are some changes in the
porosity and the strain localization equation 5.1. Note that the strain localization equation
5.1 is consistent with the second definition of Vmeso (Vmeso = Vmeso
2
).
• To consistent with Equation 5.1, it is the rectangular domain (green in Fig.5.4) should
be considered. The volume reads:
2
Vmeso = l1 l2 = 2d1 sinα(d2 + 2d1 cosα) (5.18)
• When the bounding box domain (red in Fig.5.4) is considered, the volume reads:
3
Vmeso = (l1 + d0 )(l2 + d0 ) (5.19)
In this case, Equ.5.1 should be updated with l1 + d0 and l2 + d0 instead of l1 and l2 . The
geometric compatibility (Equation 5.2) matrix A (Equ.5.7) hold.
• When the hexagonal domain (blue in Fig.5.4) is considered, the meso-volume Vmeso
h
reads,
1
Vmeso = 2d1 sinα(d2 + d1 cosα) (5.20)
A virtual rectangle with the same area of the hexagon can be considered to make the same
volume adopted in both strain localization and stress averaging processes. A rectangle of
l1∗ × l2∗ can be proposed. With this hypothesis, Equ.5.2 should be replaced by:
(
l1∗ = d2 + d1 cosα
(5.21)
l2∗ = 2d1 sinα
The matrix A needs to be updated as A∗ with two elements replaced: A∗11 = cosα and
A∗13 = −d1 sinα.
In the original version of the H-model, Vmeso is considered as the area of the hexagon,
even though this volume is not consistent with the volume l1 l2 for strain localization in
Equation 5.1. The option is inspired by the fact that granular materials in 2D can be
tessellated by contact branches.
The general scheme of the standard 2D H-model can be summarized as follows:
1. Based on given (δε1 , δε2 ), update the H-cell dimensions l1 , l2 of each H-cell according
to a kinematic localization hypothesis.
2. Update d1 , d2 , α through Equation 5.7.
3. Update N1 , N2 , T2 based on the incremental evolution of the H-cell geometry.
4. A meso-stress is defined based on the application of Love-Weber formula for each

H-cell.
5. The macroscopic stress is eventually obtained by statistical averaging of all the meso-
stresses.
In conclusion, the 2D H-model relies on three contact parameters (kn , kt and ϕg ), on

one geometrical parameter (the initial opening angle α0 ), and on αω and βω for tuning the
microstructure anisotropy.
5.1.1.4 Pathological cases
Pathological types of H-cell may be obtained owing to contact loss or gain during external
loading. There are two modes of abnormal H-cell due to contact open as shown in Fig.5.5.
The meso-stress on the broken H-cell is 0. The H-cell can be over-compressed when α ≤ 30◦
or α ≥ 90◦ , as shown in Fig.5.6. In this case, additional contacts are generated. The over-
compressed H-cells only orientate along 90◦ and for a large level of axial strain.
5.1.2 Biaxial test at the material point scale
Biaxial tests at the material point scale (homogeneous sample) are simulated using the
original version of the H-model (Nicot and Darve 2011b) based on the codes implemented
Figure 5.5: Two modes of pathological H-cell due to contact loss.
Figure 5.6: Two modes of over-compressed H-cell with the corresponding opening angle.
Table 5.1: Parameters used in biaxial tests Veylon 2017
kn (N/m) kt /kn ϕg α0 αω βω p0 (kPa)

2e8 0.5 30◦ 45◦ 0 0 100, 200, 400
1.0 0.025
100kPa 0.020
0.8 200kPa
400kPa 0.015
2)/( 1 + 2)
0.6 0.010
v
0.4 0.005
0.000
( 1
0.2
0.005
0.0 0.010
0.00 0.05 0.10 0.15 0.20 0.00 0.05 0.10 0.15 0.20
1 2 1 2
(a) (b)
Figure 5.7: The stress ratio and volumetric strain curves based on the parameters from
Veylon 2017. Three confining pressures are considered.
in Python (See in Appendix B). Under this type of loading, a constant lateral stress con-
dition is imposed together with a constant strain rate in the axial direction: δε1 = const
and δσ2 = 0. An iterative calculation process is run to find incremental strains (δε1 , δε2 )
that gives the target confining pressure. The parameters (Veylon 2017) are shown in Table
5.1. Three confining pressures 100 kPa, 200 kPa and 400 kPa are considered. The number
of directions θi is set to 360.
Stress and strain curves are presented in Fig.5.7. It can be observed that, with the
increase in p0 , there are larger maximum stress ratios and stronger dilatancy. The trend
is consistent to the results reported in Veylon 2017. If we pursue the loading up to a large
level of axial strain (ε11 was 40%), the behavior is obviously not physically relevant. As
shown in Fig.5.8, the deviatoric stress decreases to a very low value and the volumetric
strain turns to contractancy at a large level of axial strain. This is due to the fact that a
large number hexagons are broken and do not participate to stress transmission. Details
are presented in the following section.
Figure 5.8: The deviatoric stress and volumetric strain along the biaxial loading with
p0 = 100 kPa until a relative large axial strain.
5.1.3 Mesoscale inspection of the H-model during biaxial loading
The mesostructure of granular materials is always changing both in time and space. In the
H-model, directional statistics are introduced to get the average behavior at macroscale.
In each direction θi , H-cells have a specific geometry. Before introducing any additional
ingredient to the H-model, it is important to clarify the limiting aspects of the model.
The biaxial loading with p0 = 200 kPa is considered in this subsection. The evolutions
of opening angle α, the contact branches d1 and d2 and the meso-stress Vmeso σijmeso are
discussed as follows, for different H-cell directions θi .
5.1.3.1 The contact branches d1 and d2
The lengths d1 and d2 can be used to detect contact loss as shown in Fig. 5.5. The di-
rectional distribution of the ratio d1 /d0 between the length and its initial value is shown
in Fig.5.9 (a). Five specific strain states are considered as marked in Figure 5.8. The
evolution of d1 /d0 along the axial strain ε11 is presented in Fig.5.9 (b). Eight directions
ranging from 0◦ to 90◦ are considered.
0°
1.3
1.2 O
50° A
1.1 B
d1/d0 C
1.0 D
0.9
0.8
100°
150°
(a)
1.015
1.010
1.005
1.000
d1/d0
0.995
0.990
0.985 0.0° 35.0° 60.0°

15.0° 40.0° 90.0°
30.0° 45.0°
0.980
0.00 0.05 0.10 0.15 0.20 0.25 0.30 0.35 0.40
11
(b)
Figure 5.9: Evolution of the ratio between the length d1 and its initial value d0 : (a)
directional analysis at different axial strains ε11 = 0.0, 0.005, 0.032, 0.128 and 0.347 labelled
as O, A, B, C and D, respectively, (b) macro strain evolution along specific directions.
0°
1.20
1.15 O
1.10 50° A
B
1.05
d2/d0
C
1.00 D
0.95
0.90
100°
150°
(a)
1.015
1.010
1.005
1.000
d2/d0
0.995
0.990
0.985 0.0° 35.0° 60.0°

15.0° 40.0° 90.0°
30.0° 45.0°
0.980
0.00 0.05 0.10 0.15 0.20 0.25 0.30 0.35 0.40
11
(b)
Figure 5.10: Evolution of the ratio between the length d2 and its initial value d0 : (a)
directional analysis at different axial strains ε11 = 0.0, 0.005, 0.032, 0.128 and 0.347 labelled
as O, A, B, C and D, respectively, (b) macro strain evolution along specific directions.
Figure 5.11: Porosity ϕ1meso of the hexagon with respect to the opening angle α; sketch of
the two possible deforming processes.
0°
70
60 O
50 50° A
40 B
30 C
20 D
10
0
100°
150°
(a)
80
90.0° 40.0° 15.0°
60.0° 35.0° 0.0°
70 45.0° 30.0° Intial value
60
50
°
40
30
20
0.00 0.05 0.10 0.15 0.20 0.25 0.30 0.35 0.40

11
(b)
Figure 5.12: Evolution of the opening angle α: (a) directional analysis at different axial
strains ε11 = 0.0, 0.005, 0.032, 0.128 and 0.347 labelled as O, A, B, C and D, respectively,
(b) macro strain evolution along specific directions.
0°
1e7
2.5 O
2.0 50° A
meso 1.5 B
11
Vmeso C
1.0 D
0.5
0.0
100°
150°
(a)
3.0 1e7
0.0° 35.0° 60.0°
15.0° 40.0° 90.0°
2.5 30.0° 45.0°
2.0
meso
11
1.5
Vmeso
1.0
0.5
0.0
0.00 0.05 0.10 0.15 0.20 0.25 0.30 0.35 0.40
11
(b)
Figure 5.13: Evolution of Vmeso σ11meso
ε11 = 0.0, 0.005, 0.032, 0.128 and 0.347 labelled as O, A, B, C and D, respectively, (b)
macro strain evolution along specific directions.
0°
1e7
1.75 O
1.50 50° A
1.25
meso 1.00 B
22 C
0.75
Vmeso
0.50 D
0.25
0.00
100°
150°
(a)
1.8 1e7
0.0° 35.0° 60.0°
1.6 15.0° 40.0° 90.0°
30.0° 45.0°
1.4
1.2
meso
1.0
22
Vmeso
0.8
0.6
0.4
0.2
0.0
0.00 0.05 0.10 0.15 0.20 0.25 0.30 0.35 0.40
11
(b)
Figure 5.14: Evolution of Vmeso σ22meso
ε11 = 0.0, 0.005, 0.032, 0.128 and 0.347 labelled as O, A, B, C and D, respectively, (b)
macro strain evolution along specific directions.
From Fig.5.9 (a), we can observe that, from O to D (see Fig.5.8), an increasing range of
directions has d1 /d0 larger than 1, which means that more and more hexagons are broken
because of contact loss at c1 (see Fig. 5.2). At state D, there are only a small fraction
of directions around 40◦ and 140◦ with the contact c1 being unbroken. In Fig.5.9 (b), the
orientations close to 0◦ experience compression for c1 at the beginning of the loading; The
decrease in d1 /d0 is smaller and smaller along the orientation θ ranging from 0.0◦ to 90◦
in the beginning of the loading (ε < 5%); the hexagon oriented to 90◦ is the earliest one
to be broken. When θ is around 40◦ , d1 has a complex non-monotonic behavior; contacts
can re-exist after being open.
As for d2 , the curves in Fig.5.10 have some similarities and some differences to those
in Fig.5.9. Similarly, with the increase in axial strain, an increasing number H-cells have
d2 /d0 larger than 1, mainly open contacts; at the early state of the loading, the length
of d2 decreases (corresponding to compressed contacts) in the direction 0◦ , and increases
(corresponding to extended contacts) in the direction 90◦ . One important difference is that
the value of d2 /d0 oriented along 0◦ is always smaller than 1, namely the corresponding
contacts always exist during the loading.
More and more H-cells are broken around 0◦ and 90◦ because d1 or d2 extend which
results in a decrease in stress level along the loading.
5.1.3.2 The opening angle α
The value of α is closely related to the deformabitity of the mesostructure and to fabric
anisotropy. The H-cell has the potential to dilate and contract with the evolution of α as
shown in Fig.5.11.
The evolution of α is shown in Fig.5.12. It can be observed there are two directions
(around 50◦ and 130◦ ) along which α has little change over the loading. Between the two
1 in Fig.5.11; while
directions, α decreases continuously corresponding to the evolution of ○
beyond this domain it increases as the trend ○.2 The pathological case occurs resulting
from over-compression with α < 30 in the direction 90◦ .
◦
The deformed processes imply that the force chain composed by any three nearby grains
always bends along the loading direction. This is different to what has been observed in
DEM simulations where force chains tend to lie along the direction of loading after particle
rearrangements. During the rearrangements, grain loops not only are deformed, but also
are decomposed to form new loops. The generation and vanishing activities of loops is
certainly a missing puzzle piece in the H-model to account for large deformation.
5.1.3.3 The meso-stress Vmeso σijmeso
The distribution of meso-stresses Vmeso σ11meso

and Vmeso σ22
meso
in the global frame (e1 , e2 ) is
presented in Fig.5.13 - 5.14. Only unbroken hexagons are considered to contribute to the
global stresses, thus, the results are highly dependent on the length of d1 and d2 . It can be
observed that the hexagons nearby the vertical direction govern the contribution to global
stresses at the early regime. At a large level of axial strain, only a small range of directions
around 40◦ are still active, which is also observed in other revised versions of H-model, for
example the model including a shear mechanism (Zhu 2015) and the 3D H-model (Xiong
2017).
5.1.3.4 Comments
The H-model is capable of considering the deformation not only from the relative displace-
ment at the contact scale between grains, but also from the particle rearrangement through
deforming elementary hexagonal patterns of adjoining particles. The memory of history is
stored in the structure of the hexagons. Thus, the model is capable of making hardening
emerge from the deformed H-cells, and the following possible softening through the break-
age of H-cells, which is consistent with the micromechanical mechanisms of hardening and
softening observed from DEM simulations (Zhu, Nicot, and Darve 2016). Theoretically, the
model also has the potential to capture dilatancy and contractancy through the changes in
the opening angle. Making such macroscopic characteristics emerge without introducing
any empirical law at the macroscopic scale is the most significant feature of the H-model,
as a breakthrough with respect to phenomenological constitutive models.
However, the model cannot describe the critical state in a satisfying way because of
the continuous breakage of H-cells without creation of new H-cells. As investigated in the
previous chapter, the life duration of a loop is relatively short, and the generation and
vanishing process take place all along the loading though. Microstructure reorganization
in granular materials not only results in loop deformation but also induces loop transforma-
tions. Only taking into account H-cell deformation is necessary but insufficient to capture
the complex reorganization process in granular materials. Even though the hexagons are
allowed to be broken and reformed later on in the model, there is little newly-generated
hexagon after the initial state.
5.2 Emergence of critical state in the H-model
5.2.1 Method
It has been shown, in the previous chapters, that the dynamical competition between
generating and vanishing of meso-clusters always exists even at the critical state. Thus,
the H-model might be improved through introducing such a dynamical process. In order
to achieve the goal, some questions should be answered first.
• How to define the death and birth of mesostructures in the H-model?
This processes can be implemented through deactivating and reactivating H-cells. When
a H-cell is deactivated, its contacts are set to an open configuration through releasing the
overlaps at contacts c1 and c2 . Thus, the H-cell is reset with d1 = d2 = d0 . The deactivated
H-cell does not contribute to the global stresses at this loading step. During the reset, the
volume of the H-cell is kept constant by adjusting the opening angle accordingly. In real
granular materials, the contact network rearranges instantaneously resulting in different
meso-structures with constant volume.
In granular materials, a sample under loading evolves and adapts through grain rear-
rangements. The fabric of the assembly evolves to be coincident with the loading direction
(Li and Li 2009). Based on this fact, the reactivated H-cells could be adjusted from the
old configuration. In the H-model, the fabric is mainly controlled by the opening angles α
at mesoscale and the distribution function ω(θ) at the macroscale. To work at mesoscale,
a scheme for resetting α is proposed for reactivated hexagons with the porosity being kept,
as shown in Fig.5.15. In the resetting process, the shape of the hexagon is stretched along
the compressed direction.
• Which hexagons should be deactivated/reactivated?
As a first step, we propose to conduct the deactivation/reactivation scheme on the H-cells

that are chosen randomly at each loading step. The percentage ph of hexagons renewed
over an axial strain increment δε11 can be set according to the renewal rate ξ characterized
by the axial strain. Thus, ph = δεξ1 . As obtained in the previous chapter, the renewal rate
of L6p at critical state is ranging from 0.3% to 2.79% as shown in Section 4.4.
5.2. Emergence of critical state in the H-model 115
Figure 5.15: Schematic model to reset α with the porosity being kept. The reset 1 is
triggered when the opening angle smaller than 60◦ tends to decrease further, and the reset
2 when the opening angle larger than 60◦ tends to increase further.
Random reset has been used in complex chaotic system in physics. It has been proven
by Chen 1996 in chaotic systems, that if their velocities are randomly reset at a regular time
interval ∆t according to a prescribed probability distribution, then it might happen that
all the particles will eventually converge to some unique yet erratic orbit in the physical
space, depending on the system style. Similar convergent effect should be expected on the
H-model after randomly resetting the hexagons.
5.2.2 Preliminary results
A biaxial test is predicted using the updated H-model. The flowchart of the computation
incorporating the deactivation/reactivation process is shown in Fig. 5.16. One confining
pressure p0 = 100 kPa is considered. The renewal rate ξ is set to 1.5%. Parameters are
shown in Table 5.1. The simulated deviatoric stress (q = σ11 − σ22 ) and volumetric strain
(εv = ε11 + ε22 ) are presented in Fig. 5.17. It can be observed that the stress-strain curves
tend to reach a stationary state with fluctuations. Note that it is the volume Vmeso
2
consid-
ered in both strain localization and stress averaging processes, as noted in green in Fig.5.4.
The directional distribution of geometrical variables d1 /d0 , d2 /d0 and α of the H-cell
is analyzed at four different strain levels ε11 = 0.000, 0.005, 0.150 and 0.350 labelled as
Figure 5.16: Flowchart of the computation of biaxial loading incorporating the deactiva-
tion/reactivation scheme.
Figure 5.17: Deviatoric stress and volumetric strain with respect to the axial strain using
the updated H-model.
0° 0°
1.020 1.020
1.015 O 1.015 A
1.010 50° 1.010 50°
1.005 1.005
1.000 1.000
d1/d0
d1/d0
0.995 0.995
0.990 0.990
0.985 0.985
0.980 0.980
100° 100°
150° 150°
0° 0°
1.04 B 1.04 C
1.02 50° 1.02 50°
1.00 1.00
d1/d0
d1/d0
0.98 0.98
0.96 0.96
100° 100°
150° 150°
Figure 5.18: Polar diagrams of the ratio between the length d1 and its initial value d0 :
at different axial strains ε11 = 0.000, 0.005, 0.150 and 0.350 labelled as O, A, B and C,
respectively (see Fig.5.17).
0° 0°
1.020 1.020
1.015 O 1.015 A
1.010 50° 1.010 50°
1.005 1.005
1.000 1.000
d2/d0
d2/d0
0.995 0.995
0.990 0.990
0.985 0.985
0.980 0.980
100° 100°
150° 150°
0° 0°
1.04 B 1.04 C
1.02 50° 1.02 50°
1.00 1.00
d2/d0
d2/d0
0.98 0.98
0.96 0.96
100° 100°
150° 150°
Figure 5.19: Polar diagrams of the ratio between the length d2 and its initial value d0 :
at different axial strains ε11 = 0.000, 0.005, 0.150 and 0.350 labelled as O, A, B and C,
respectively.
0° 0°
70 70
60 O 60 A
50 50° 50 50°
40 40
30 30
20 20
10 10
0 0
100° 100°
150° 150°
0° 0°
70 70
60 B 60 C
50 50° 50 50°
40 40
30 30
20 20
10 10
0 0
100° 100°
150° 150°
Figure 5.20: Polar diagrams of the opening angle α: at different axial strains ε11 = 0.000,
0.005, 0.150 and 0.350 labelled as O, A, B and C, respectively (see Fig. 5.17).
5.3. Conclusions 121
600 0.05
40° 40°
500 45° 0.04 45°
400 0.03
q (kPa)
300 0.02
v
200 0.01
100 0.00
0 0.01
0.00 0.05 0.10 0.15 0.20 0.25 0.30 0.35 0.40 0.00 0.05 0.10 0.15 0.20 0.25 0.30 0.35 0.40
11 11
Figure 5.21: Deviatoric stress and volumetric strain with different initial opening angles
40◦ and 45◦ being considered, predicted by the updated H-model.
O, A, B and C , respectively, in Fig. 5.17. Results are presented in Fig. 5.18 - 5.20. At
the beginning stage O, there are four randomly-selected directions along which the H-cells
have open contacts with d1 /d0 = d2 /d0 = 1. The reset scheme for α is triggered in one
direction among the four, presented as a sharp angle. At the strain level A, open contacts
due to d2 /d0 > 1 exist along a range of directions around 90◦ . Among the similar range of
directions, a number of opening angles have been reset. At the stationary stage, the curves
at stage B and stage C are quite similar, which means that the two configurations are alike.
It is believed that this is the reason why a stationary state emerge can be obtained at the
macroscopic scale in the prediction from the deactivation and reactivation scheme.
An additional simulation with a smaller initial opening angle (40◦ ) has been conducted
as shown in Fig.5.21. It can be observed in Fig.5.21 that the decrease in the opening angle
results in a higher q peak, a larger softening, and stronger dilatancy. This is consistent with
the fact that a denser sample has a larger initial stiffness and stronger dilatancy. When
the initial angle decreases from 45◦ to 40◦ , the porosity decreases as shown in Fig. 5.11.
5.3 Conclusions
The H-model is capable of considering the deformation not only from contact displacements
but also from particle rearrangements through deformed H-cells. It is able to make the
macromechanical characteristic of granular materials emerge, e.g.dilatancy/contractancy
and hardening/softening, without introducing any empirical law at macroscopic scale.
Following the spirit of directional distribution in the H-model, a method is proposed to

make the critical state emerge through a deactivation/reactivation procedure of hexagons
based on the dynamical view of the critical state in Chapter 4. It has been shown that the
deviatoric stress and volumetric strain along a biaxial test reach a steady state. Directional
analysis has been conducted. It is demonstrated that the deactivation/reactivation scheme
can help to restrict the continuously increasing number of opening contacts. The updated
H-model seems to have the ability to generate similar structural configurations leading to
a stationary state at the macroscopic scale.
Further efforts need to be devoted to polish and precise the deactivation/reactivation

scheme through analyzing the sensitivity of the parameters. Especially for the mesostruc-
tural reorganization rate ξ, the value is inconstant before reaching the critical state accord-
ing to the results in the previous chapter. Detailed comparisons like anisotropy between
DEM simulations and the modeling in this chapter can help to further evaluate the model.
In addition, prediction of experimental data using the H-model is worth putting forward
in the future.
Chapter 6
General conclusion and perspectives
Contents
6.1 Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 123
6.1.1 Dilatant proportional strain loading and relationships with critical state123
6.1.2 Dynamical view of critical state . . . . . . . . . . . . . . . . . . . . . 124
6.1.3 Enriched H-model with emerging critical state . . . . . . . . . . . . . 125
6.2 Perspectives . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 125
6.2.1 Micromechanical investigations . . . . . . . . . . . . . . . . . . . . . 125
6.2.2 H-model . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 128
6.2.3 Multiscale modeling of landslides . . . . . . . . . . . . . . . . . . . . 129
6.1 Conclusion
This PhD work was devoted to micromechanical investigations and multiscale modeling of
granular materials with an emphasise on the critical state concept.
6.1.1 Dilatant proportional strain loading and relationships with

critical state
A relation has been established between the mechanical states reached along a proportional
strain test and the critical states obtained from biaxial tests within the framework of ACST.
Macroscopic responses and fabric-related measures in homogeneous domains along biax-
ial tests converge towards the evolution curve of those variables from proportional strain
tests at critical state. Given these results, only one single test is therefore necessary to
construct the classical critical state line in (p - q - e∗ - αc∗ - ri∗ ) space for any granular
123
124 Chapter 6. General conclusion and perspectives
material. Therefore, the CS concept can readily be obtained for a wide class of loading
paths. Indeed, CS acts as a general attractor irrespective of the loading path considered.
The mixed proportional strain and biaxial loading paths further confirm this relation. A
more general definition of CS can be given as follows: a p − q − e state, for a sample under
constant stress and strain rate directions, after a given loading history, is considered as a
critical state if the application of a biaxial loading under the same lateral pressure starting
from this p − q − e state leads to zero volume strain and zero deviatoric stress evolution. In
addition, the comparison between the imposed dilatancy/contractancy along proportional
strain paths and unconstrained dilatancy/contractancy expected along biaxial paths pro-
vides valuable information to interpret the various loading paths obtained in p − q plane
for proportional strain loading paths.
6.1.2 Dynamical view of critical state
We have shown that the critical state results from imperfect detailed balance competing
between generating and vanishing of meso-clusters. We have established that force chains
and grain loops have a regular, short life duration (0.4-3% compared to the whole 15%
loading in a biaxial test) at critical state, corresponding to a rapid fading of the material
memory along its loading history. Critical state "forgot" not just the initial microstructure
conformation but its recent past in the stationary regime in a sample under continuous
shearing. We have shown that the uniqueness of the critical state that holds for statis-
tical descriptors also holds for the underlying dynamical processes. By fitting the data,
we have shown that force chains and loops of different sizes have different lifespan and
life expectancy which increase with the mean stress level. Larger loops have a shorter life
due to their deformability and force chain life duration is similar to the life duration of
small order loops. This is consistent with the known relationship between force chains and
their small order supporting loops. Moreover, the microstructure reorganization has been
proved to rely on two mechanisms that act over very different axial strain increments. A
local mechanism related to sheared mesostructures provokes a rapid (i.e. for small ax-
ial strain increments) reorganization of the microstructure while the characteristic strain
related to the nucleation of new sheared mesostructured accounts for the microstructure
reorganization over larger axial strain increments.
6.2. Perspectives 125
6.1.3 Enriched H-model with emerging critical state
Following the spirit of directional distribution in the H-model, a method is proposed to ap-
proach the microstructural mechanisms taking place when approaching the critical state,
by introducing a deactivation/reactivation procedure on H-cells, based on DEM results
about the dynamical view of critical state. It has been shown that the deviatoric stress
and volumetric strain along a biaxial test reach a steady state. It is demonstrated that
the deactivation/reactivation scheme can help to limit contact loss and occurrence of other
pathological cases in the deformation of H-cells. The updated H-model seems to have the
ability to generate similar structural configurations leading to a stationary regime at the
macroscopic scale in biaxial tests which corresponds to critical state.
6.2 Perspectives
6.2.1 Micromechanical investigations

• Bursts of kinetic energy and global reorganization
Studies proved that outbursts of kinetic energies can be seen as signatures of mechanical
instabilities and inertial transitions in granular materials (Gaume, Chambon, and Naaim
2011; Wautier, Bonelli, and Nicot 2018b; Clerc et al. 2021). While evolving at critical
state, these bursts were shown to have a finite size (Clerc et al. 2021), and inertial transi-
tions are limited in space and time. Following the results on the two mechanisms forcing
microstructure reorganization in this thesis, it will be interesting to give a close look to
outbursts of kinetic energy, and to characterize the reorganization rate within the domain
of the bursts using the lifespan PDF. So that we can further refine how successive local
inertial events can result in a global microstructure reorganization. Investigations on the
evolution of grain loops and force chains at the critical state and during bursts of kinetic
energy can help to improve the understanding of the co-evolution of the structures and
stresses.
• Further clarify the mechanism of critical state through studying transition of confor-
mations
In this thesis, using PDFs of lifespan and life expectancy of mesostructures, we character-
ized the dynamical reorganization rates in granular materials during loading. The vanished
400
100kPa-dense ( = 0.6)
100kPa-dense ( = 0.7)
350 100kPa-dense ( = 0.8)
300 100kpa-dense ( 0 = 100kPa)
Maximum stress ratio
250 Critical stress ratio C
q (kPa)
200
150
B
100 A
50
00 50 100 150 200 250 300 350 400

p (kPa)
Figure 6.1: Stress paths in p − q plane along triaxial tests and proportional strain tests
with different dilatant rates characterized by κ (dεv = (1 + 2κ)dε2 ).
mesostructures are transferred to other configuration. At local scale, we can track the move
of each particle and explore the transitions of conformations and the possibility of reform-
ing disappeared mesostructures. This could help to understand the leakage or escapes
of the imperfect detailed balance (as mentioned in Chapter 4) as well as memory fading
process during different loading conditions in granular materials, thus further clarify the
mechanism of the critical state as an attractor.
• Further verify the conclusions obtained in the thesis in 3D situation
It will be interesting to further verify the conclusions in this thesis in 3D. Some preliminary
results in 3D about dilatant proportional strain loading and their relationship to critical
state are shown in Fig.6.1 and 6.2. Three triaxial tests were simulated to obtain standard
critical states. Three dilatant proportional tests within different dilatant rates character-
ized by κ are considered with dεv = (1 + 2κ)dε2 . A "stress loop" can be obtained when
κ = −0.7 in Fig.6.1.
A complex loading mixing a proportional dilatant strain test and a triaxial test with
350
v of proportional strain loading q of proportional strain loading
v of continued triaxial loading q of continued triaxial loading 0.08
300
250 0.06
200
q (kPa)
0.04
v
150
100 0.02
50
0.00
0 0.00 0.05 0.10 0.15 0.20 0.25
2
Figure 6.2: Stress-strain responses along mixed proportional strain and triaxial loading
paths in 3D. At the transition state, the lateral stress is kept unchanged. The loading of
100kPa-dense (κ = −0.7) as shown in Fig.6.1 is adopted here. ε̇v = (1 + 2κ)ε̇1
the lateral stress kept unchanged has been simulated in 3D and the corresponding stress-
strain responses are presented in Fig. 6.2. The transition points are marked by dots. It
can be observed that εv and q along the continued triaxial test tend to be stable. This
result suggests the possibility of extending the generalized definition of critical state to 3D
conditions. More efforts are worth being devoted into micromechanical investigations in
the 3D simulations on force chains, and grain loops benefiting from the concept of grain
loops that was extended to 3D conditions recently (Chueire et al. n.d.).
The concept of lifespan and life expectancy presented in the thesis should be valid to
any types of meso-structure well defined in both 2D and 3D conditions. However, the
amplitude of the reorganization rate of mesostructures in 3D are expected to be different.
• Refine the results using angular particles
In this thesis, spherical particles was used. Simulations with angular particles merit at-
tention in future studies. The real shape of particles can be simulated using Level-Set
(LS-DEM) shape description through avatar paradigm and X-ray computed tomographic
(XRCT) images (Kawamoto et al. 2018; Duriez and Galusinski 2021). In a system with
angular particles, the reorganization rate would be different since the angularity has effect
on the rolling at contacts. We expect that the mesostructures will live longer for angular
particles.
6.2.2 H-model
• Further improve the deactivation/reactivation scheme.
The deactivation/reactivation scheme can be further improved from different perspectives.

The volume of H-cell was kept constant during the reset operation through adjusting the
opening angle α, namely Vmeso adjust before
= Vmeso . There are other possible options to reset the
adjust before
Vmeso −Vmeso
H-cell, for example, by controlling before
Vmeso 22 . A full implementation of the
= δε11 +δεtry
life expectancy PDF in the selection of the H-cells to reset may also improve the procedure.
The relation of the material porosity, the loading rate and the confining pressure etc. to
the structural reorganization rate ξ can be further refined. Another potential direction
to improve the scheme is to find the limit condition of a loop in a sample from DEM
simulations first, and then control the deactivation and reactivation of each H-cell according
to the configuration itself.
• Introduce the deactivation/reactivation scheme into the 3D H-model.
The H-model has been extended to 3D with mesostructures of ten grains consisting of two
imbricated hexagons in perpendicular directions (Xiong 2017). It would be interesting to
introduce the deactivation/reactivation scheme into the 3D H-model.
6.2.3 Multiscale modeling of landslides
The H-model has been implemented in different computational frameworks to simulate a

variety of geotechnical problems where multi-physical couplings can occur (Xiong, Yin, and
Nicot 2019; Xiong et al. 2021; Wautier et al. 2021). The H-model has been coupled with
Finite Element Method (FEM), based on which, a strip footing problem in plane strain
conditions has been analyzed as a non-homogeneous boundary value problem (Xiong, Yin,
and Nicot 2019). The H-model has been implemented into a Finite Difference Method
(FDM) solver to perform a stress analysis for an earth dam as the upstream water level
gradually increases (Wautier et al. 2021). The model has been used with smoothed particle
hydrodynamics (SPH) method to simulate collapse of granular material (Xiong et al. 2021).
One collapse is simulated based on the H-model and SPH method, as presented in Figure
6.3. The multiscale approach using the H-model is helpful in improving the understanding
of non-homogeneous boundary value problems and should be an effective computational
tool for the analysis of geotechnical boundary value problems.
There will be great interest to apply the improved version of H-model developed in
this thesis coupling with a numerical platform, for example, SPH method to landslide
modeling. The Bas-Verger landslide in the Pays d'Auge plateau (Normandy, France) could
be an appropriate case to start with (Fig. 6.4). The Bas-Verger landslide is a shallow
planar slide. The landslide surface morphology is attainable from previous studies based
on field mapping and aerial image interpretation (Fressard et al. 2016). It is promising
to carry out simulations along cross-sections to analyze the potential triggering processes
of the landslide. Comparisons between results from multiscale modeling and from a more
classical approach used for landslide hazard analysis can be helpful to further demonstrate
the interest of the new approach as well as the pros and cons of each method.
(a)
(b)
Figure 6.3: simulation of the collapse of granular column using SPH particles and the H-
model: (a) initial configuration and (b) color map of the deformation in direction 1 after
collapse.
Figure 6.4: Aerial view of the Bas-Verger landslid in 2006 with associated localization of
the field measurements conducted on the landslide (Fressard et al. 2016).
Appendix A
Complementary results on critical state

dynamics
In addition to the results presented in the main part of the chapter, figures A.1 and
A.2 provides complementary results on a) the detailed balance achieved at critical state
for varying confining pressures and b) the uniqueness of the PDF for lifespan and life
expectancy in dense and loose samples provided the strain rate is rescaled in the dense
case to account for strain localization.
133
134 Appendix A. Complementary results on critical state dynamics
200
40kPa-loose
175 40kPa-dense*
150
125
100
PDF
75
50
25
0
0.04 0.03 0.02 0.01 0.00 0.01 0.02 0.03 0.04
(a) 2
100
400kPa-loose
400kPa-dense*
80
60
PDF
40
20
0
0.04 0.03 0.02 0.01 0.00 0.01 0.02 0.03 0.04
(b) 2
Figure A.1: PDFs for chained particles under two confining pressures: (a) 40kPa, PDFs
from the 40Kpa-dense are stretched by strain magnitude ratio 1.9; (b) 400kPa, the 400kPa-
dense stretched by 3.2.
135
Loop 3 Loop 4
200 200
40kPa-loose
175 40kPa-dense* 175
150 150
125 125
100 100
PDF
75 75
50 50
25 25
0 0
0.04 0.02 0.00 0.02 0.04 0.04 0.02 0.00 0.02 0.04
Loop 5 Loop 6p
200 200
175 175
150 150
125 125
100 100
PDF
75 75
50 50
25 25
0 0
0.04 0.02 0.00 0.02 0.04 0.04 0.02 0.00 0.02 0.04
2 2
(a)
Loop 3 Loop 4
200 200
400kPa-loose
175 400kPa-dense* 175
150 150
125 125
100 100
PDF
75 75
50 50
25 25
0 0
0.04 0.02 0.00 0.02 0.04 0.04 0.02 0.00 0.02 0.04
Loop 5 Loop 6p
200 200
175 175
150 150
125 125
100 100
PDF
75 75
50 50
25 25
0 0
0.04 0.02 0.00 0.02 0.04 0.04 0.02 0.00 0.02 0.04
2 2
(b)
Figure A.2: PDFs for grain loops under two confining pressures: (a) 40kPa, PDFs from the
40Kpa-dense are stretched by strain magnitude ratio 1.9; (b) 400kPa, the 400kPa-dense
stretched by 3.2.
Bibliography
Antony, SJ (2007). “Link between single-particle properties and macroscopic properties in

particulate assemblies: role of structures within structures.” In: Philosophical Transac-
tions of the Royal Society A: Mathematical, Physical and Engineering Sciences 365.1861,
pp. 2879–2891 (cit. on p. 56).
Azema, Emilien, Farhang Radjai, and Jean-Noël Roux (2015). “Internal friction and ab-
sence of dilatancy of packings of frictionless polygons.” In: Physical Review E 91.1,
p. 010202 (cit. on p. 7).
Babu, Varghese et al. (2020). “Is friction essential for dilatancy and shear jamming in
granular matter?” In: arXiv preprint arXiv:2003.08815 (cit. on p. 7).
Been, K, Mike G Jefferies, and J Hachey (1991). “Critical state of sands.” In: Géotechnique
41.3, pp. 365–381 (cit. on pp. 2, 10, 23, 56).
Been, Ken and Mike G Jefferies (1985). “A state parameter for sands.” In: Géotechnique
35.2, pp. 99–112 (cit. on pp. 10, 23).
Behringer, Robert P and Bulbul Chakraborty (2018). “The physics of jamming for granular
materials: a review.” In: Reports on Progress in Physics 82.1, p. 012601 (cit. on p. 55).
Ben-Nun, Oded, Itai Einav, and Antoinette Tordesillas (2010). “Force attractor in confined
comminution of granular materials.” In: Physical review letters 104.10, p. 108001 (cit.
on p. 56).
Bolton, MD (1986). “The strength and dilatancy of sands.” In: Geotechnique 36.1, pp. 65–
78 (cit. on p. 7).
Bonelli, Stéphane et al. (2012). “On the definition of an average strain tensor for two-
dimensional granular material assemblies.” In: International Journal of Solids and Struc-
tures 49.7-8, pp. 947–958 (cit. on p. 99).
Casagrande, Arthur (1936). “Characteristics of cohesionless soils affecting the stability of
slopes and earth fills.” In: Journal of the Boston Society of Civil Engineers 23.1, pp. 13–
32 (cit. on pp. 2, 10, 56).
Chang, CS and P-Y Hicher (2005). “An elasto-plastic model for granular materials with
microstructural consideration.” In: International journal of solids and structures 42.14,
pp. 4258–4277 (cit. on p. 20).
Chen, Yih-Yuh (1996). “Why do chaotic orbits converge under a random velocity reset?”
In: Physical review letters 77.21, p. 4318 (cit. on p. 115).
Chu, J, S CR Lo, and Ian Kenneth Lee (1992). “Strain-softening behavior of granular soil
in strain-path testing.” In: Journal of Geotechnical Engineering 118.2, pp. 191–208 (cit.
on pp. 8, 31, 34).
Chueire, J et al. (n.d.). “On the extension of the grain loop concept from 2d to 3d granular
assemblies (Under review).” In: () (cit. on p. 128).
137
138 Bibliography
Clavaud, Cécile et al. (2017). “Revealing the frictional transition in shear-thickening sus-
pensions.” In: Proceedings of the National Academy of Sciences 114.20, pp. 5147–5152
(cit. on p. 7).
Clerc, A et al. (2021). “Meso-scale signatures of inertial transitions in granular materials.”
In: Granular Matter 23.2, pp. 1–14 (cit. on p. 125).
Cundall, Peter A (1971). “A computer model for simulating progressive, large-scale move-
ment in blocky rock system.” In: Proceedings of the International Symposium on Rock
Mechanics, 1971 (cit. on p. 13).
Cundall, Peter A and Otto DL Strack (1979). “A discrete numerical model for granular
assemblies.” In: Géotechnique 29.1, pp. 47–65 (cit. on pp. 13, 14, 24, 25).
Da Cruz, Frédéric et al. (2005). “Rheophysics of dense granular materials: Discrete simu-
lation of plane shear flows.” In: Physical Review E 72.2, p. 021309 (cit. on p. 26).
Daouadji, Ali et al. (2013). “Experimental and numerical investigation of diffuse instability
in granular materials using a microstructural model under various loading paths.” In:
Géotechnique 63.5, pp. 368–381 (cit. on pp. 31, 34).
Daouadji, Ali et al. (2017). “Phase transformation states of loose and dense granular ma-
terials under proportional strain loading.” In: Journal of Engineering Mechanics 143.1,
p. C4016007 (cit. on pp. 24, 34, 52).
Darve, Félix (1996). “Liquefaction phenomenon of granular materials and constitutive sta-
bility.” In: Engineering Computations: Int J for Computer-Aided Engineering 13.7,
pp. 5–28 (cit. on p. 52).
Darve, Félix et al. (2020). “Slip lines versus shear bands: two competing localization
modes.” In: Mechanics Research Communications, p. 103603 (cit. on pp. 79, 93).
Deng, Na et al. (2021a). “Dynamic view of critical state regime in granular materials: a
mesoscale perspective.” In: EPJ Web of Conferences. Vol. 249. EDP Sciences, p. 11011
(cit. on p. 57).
Deng, Na et al. (2021b). “Lifespan dynamics of cluster conformations in stationary regimes
in granular materials (in revision).” In: Physical Review E (cit. on p. 3).
Deng, Na et al. (2021c). “On the attraction power of critical state in granular materials.”
In: Journal of the Mechanics and Physics of Solids 149, p. 104300 (cit. on pp. 3, 56).
Desrues, J et al. (1996). “Void ratio evolution inside shear bands in triaxial sand specimens
studied by computed tomography.” In: Géotechnique 46.3, pp. 529–546 (cit. on pp. 28,
38).
Desrues, Jacques et al. (2019). “From discrete to continuum modelling of boundary value
problems in geomechanics: An integrated FEM-DEM approach.” In: International Jour-
nal for Numerical and Analytical Methods in Geomechanics 43.5, pp. 919–955 (cit. on
p. 20).
Drescher, A and G De Josselin De Jong (1972). “Photoelastic verification of a mechanical
model for the flow of a granular material.” In: Journal of the Mechanics and Physics of
Solids 20.5, pp. 337–340 (cit. on pp. 10, 16, 56, 59).
Bibliography 139
Drucker, Daniel C, Robert E Gibson, and David J Henkel (1957). “Soil mechanics and
work-hardening theories of plasticity.” In: Transactions of the American Society of Civil
Engineers 122.1, pp. 338–346 (cit. on p. 8).
Duriez, Jérôme and Cédric Galusinski (2021). “A Level Set-Discrete Element Method in
YADE for numerical, micro-scale, geomechanics with refined grain shapes.” In: Com-
puters & Geosciences 157, p. 104936 (cit. on p. 128).
Fish, Jacob, Gregory J Wagner, and Sinan Keten (2021). “Mesoscopic and multiscale mod-
elling in materials.” In: Nature Materials 20.6, pp. 774–786 (cit. on p. 1).
Fressard, Mathieu et al. (2016). “Multi-method characterisation of an active landslide:
Case study in the Pays d’Auge plateau (Normandy, France).” In: Geomorphology 270,
pp. 22–39 (cit. on pp. 129, 131).
Fu, Pengcheng and Yannis F Dafalias (2011). “Fabric evolution within shear bands of
granular materials and its relation to critical state theory.” In: International Journal
for Numerical and Analytical Methods in Geomechanics 35.18, pp. 1918–1948 (cit. on
pp. 9, 10, 16, 23, 26, 28, 41, 56, 59).
— (2015). “Relationship between void- and contact normal- based fabric tensors for 2D
idealized granular materials.” In: International Journal of Solids and Structures 63,
pp. 68–81 (cit. on pp. 10, 23, 26, 28, 39, 41).
Gao, Zhiwei et al. (2014). “A critical state sand plasticity model accounting for fabric evolu-
tion.” In: International Journal for Numerical and Analytical Methods in Geomechanics
38.4, pp. 370–390 (cit. on p. 8).
Gardner, Mark R and W Ross Ashby (1970). “Connectance of large dynamic (cybernetic)
systems: critical values for stability.” In: Nature 228.5273, pp. 784–784 (cit. on pp. 2,
55).
Gaume, Johan, Guillaume Chambon, and Mohamed Naaim (2011). “Quasistatic to inertial
transition in granular materials and the role of fluctuations.” In: Physical Review E 84.5,
p. 051304 (cit. on p. 125).
Guo, Ning (2014). “Multiscale characterization of the shear behavior of granular media.”
PhD thesis (cit. on p. 20).
Hettler, A and I Vardoulakis (1984). “Behaviour of dry sand tested in a large triaxial
apparatus.” In: Geotechnique 34.2, pp. 183–197 (cit. on p. 9).
Hill, Rodney (1963). “Elastic properties of reinforced solids: some theoretical principles.”
In: Journal of the Mechanics and Physics of Solids 11.5, pp. 357–372 (cit. on p. 13).
Houlsby, Guy Tinmouth (1991). “How the dilatancy of soils affects their behaviour.” In:
(cit. on p. 6).
Ibraim, E et al. (2010). “Strain path controlled shear tests on an analogue granular mate-
rial.” In: Géotechnique 60.7, pp. 545–559 (cit. on pp. 24, 31, 34).
Imole, Olukayode I et al. (2014). “Micro-macro correlations and anisotropy in granular
assemblies under uniaxial loading and unloading.” In: Physical Review E 89.4, p. 042210
(cit. on p. 16).
140 Bibliography
Jaeger, Heinrich M, Sidney R Nagel, and Robert P Behringer (1996). “Granular solids,
liquids, and gases.” In: Reviews of modern physics 68.4, p. 1259 (cit. on p. 1).
Kawamoto, Reid et al. (2018). “All you need is shape: predicting shear banding in sand
with LS-DEM.” In: Journal of the Mechanics and Physics of Solids 111, pp. 375–392
(cit. on pp. 10–12, 128).
Kruyt, Nicolaas P (2003). “Statics and kinematics of discrete Cosserat-type granular ma-
terials.” In: International journal of solids and structures 40.3, pp. 511–534 (cit. on
p. 18).
Kruyt, Nicolaas P and L Rothenburg (2014). “On micromechanical characteristics of the
critical state of two-dimensional granular materials.” In: Acta Mechanica 225.8, pp. 2301–
2318 (cit. on pp. 10, 15, 16, 19, 26, 56).
— (2016). “A micromechanical study of dilatancy of granular materials.” In: Journal of
the Mechanics and Physics of Solids 95, pp. 411–427 (cit. on pp. 6, 7, 10, 25).
Kruyt, NP and L Rothenburg (1996). “Micromechanical definition of the strain tensor for
granular materials.” In: (cit. on p. 18).
Kuhn, Matthew R (1999). “Structured deformation in granular materials.” In: Mechanics
of materials 31.6, pp. 407–429 (cit. on p. 18).
— (2016a). “Maximum disorder model for dense steady-state flow of granular materials.”
In: Mechanics of Materials 93, pp. 63–80 (cit. on p. 57).
— (2016b). “The critical state of granular media: convergence, stationarity and disorder.”
In: Géotechnique 66.11, pp. 902–909 (cit. on pp. 10, 15, 57, 63).
Li, X S (1997). “Modeling of dilative shear failure.” In: Journal of Geotechnical and Geoen-
vironmental Engineering 123.7, pp. 609–616 (cit. on p. 38).
Li, X S and YF Dafalias (2015). “Dissipation consistent fabric tensor definition from DEM
to continuum for granular media.” In: Journal of the Mechanics and Physics of Solids
78, pp. 141–153 (cit. on pp. 10, 46).
Li, Xia and Xiang Song Li (2009). “Micro-macro quantification of the internal structure of
granular materials.” In: Journal of Engineering Mechanics 135.7, pp. 641–656 (cit. on
p. 114).
Li, Xiang Song and Yannis F Dafalias (2000). “Dilatancy for cohesionless soils.” In: Géotech-
nique 50.4, pp. 449–460 (cit. on pp. 6, 7).
— (2012). “Anisotropic critical state theory: role of fabric.” In: Journal of Engineering
Mechanics 138.3, pp. 263–275 (cit. on pp. 8–11, 20, 21, 39, 50).
Li, Xiang Song, Yannis F Dafalias, and Zhi Liang Wang (1999). “State dependant dilatancy
in critical-state constitutive modelling of sand.” In: Canadian Geotechnical Journal
36.4, pp. 599–611 (cit. on p. 38).
Liang, Weijian and Jidong Zhao (2019). “Multiscale modeling of large deformation in ge-
omechanics.” In: International Journal for Numerical and Analytical Methods in Ge-
omechanics 43.5, pp. 1080–1114 (cit. on p. 1).
Bibliography 141
Liu, Jiaying, François Nicot, and Wei Zhou (2018). “Sustainability of internal structures
during shear band forming in 2D granular materials.” In: Powder Technology 338,
pp. 458–470 (cit. on pp. 19, 28, 59, 63).
Liu, Jiaying et al. (2020a). “Macroscopic softening in granular materials from a mesoscale
perspective.” In: International Journal of Solids and Structures (cit. on pp. 9, 26, 29,
55, 82).
Liu, Jiaying et al. (2020b). “Strong contacts, connectivity and fabric anisotropy in granular
materials: A 3D perspective.” In: Powder Technology (cit. on p. 29).
Luding, Stefan, Yimin Jiang, and Mario Liu (2021). “Un-jamming due to energetic insta-
bility: statics to dynamics.” In: Granular Matter 23.4, pp. 1–41 (cit. on p. 18).
Majmudar, Trushant S and Robert P Behringer (2005). “Contact force measurements and
stress-induced anisotropy in granular materials.” In: Nature 435.7045, pp. 1079–1082
(cit. on p. 12).
Majmudar, TS et al. (2007). “Jamming transition in granular systems.” In: Physical review
letters 98.5, p. 058001 (cit. on p. 56).
Matsuoka, Hajime (1974). “Dilatancy characteristics of soil.” In: Soils and foundations 14.3,
pp. 13–24 (cit. on p. 7).
Matsushima, Takashi and Raphael Blumenfeld (2014). “Universal structural characteristics
of planar granular packs.” In: Physical review letters 112.9, p. 098003 (cit. on p. 55).
— (2017). “Fundamental structural characteristics of planar granular assemblies: Self-
organization and scaling away friction and initial state.” In: Physical Review E 95.3,
p. 032905 (cit. on p. 81).
May, Robert M (1972). “Will a large complex system be stable?” In: Nature 238.5364,
pp. 413–414 (cit. on pp. 2, 55).
Meier, HA, P Steinmann, and E Kuhl (2009). “On the multiscale computation of con-
fined granular media.” In: ECCOMAS Multidisciplinary Jubilee Symposium. Springer,
pp. 121–133 (cit. on p. 20).
Milo, Ron et al. (2002). “Network motifs: simple building blocks of complex networks.” In:
Science 298.5594, pp. 824–827 (cit. on p. 55).
Muthuswamy, Maya and Antoinette Tordesillas (2006). “How do interparticle contact fric-
tion, packing density and degree of polydispersity affect force propagation in particu-
late assemblies?” In: Journal of Statistical Mechanics: Theory and Experiment 2006.09,
P09003 (cit. on p. 17).
Nguyen, Ngoc-Son, Hélène Magoariec, and Bernard Cambou (2012). “Local stress analysis
in granular materials at a mesoscale.” In: International Journal for Numerical and
Analytical Methods in Geomechanics 36.14, pp. 1609–1635 (cit. on pp. 18, 19).
Nguyen, Ngoc-Son et al. (2009). “Analysis of structure and strain at the meso-scale in 2D
granular materials.” In: International journal of solids and structures 46.17, pp. 3257–
3271 (cit. on pp. 18, 19).
142 Bibliography
Nicot, François and Félix Darve (2007). “Basic features of plastic strains: from micro-
mechanics to incrementally nonlinear models.” In: International Journal of Plasticity
23.9, pp. 1555–1588 (cit. on p. 52).
— (2011a). “Diffuse and localized failure modes: two competing mechanisms.” In: Interna-
tional Journal for Numerical and Analytical Methods in Geomechanics 35.5, pp. 586–
601 (cit. on pp. 27, 28).
— (2011b). “The H microdirectional model: accounting for a mesoscopic scale.” In: Me-
chanics of Materials 43.12, pp. 918–929 (cit. on pp. 3, 20, 21, 26, 56, 93, 97, 99, 102).
Nicot, François, Luc Sibille, and Pierre-Yves Hicher (2015). “Micro–macro analysis of gran-
ular material behavior along proportional strain paths.” In: Continuum Mechanics and
Thermodynamics 27.1-2, pp. 173–193 (cit. on pp. 31, 34).
Nicot, François et al. (2013a). “Granular media failure along triaxial proportional strain
paths.” In: European Journal of Environmental and Civil Engineering 17.9, pp. 777–790
(cit. on p. 34).
Nicot, François et al. (2013b). “On the definition of the stress tensor in granular media.” In:
International Journal of Solids and Structures 50.14-15, pp. 2508–2517 (cit. on pp. 29,
99).
Oda, M and H Kazama (1998). “Microstructure of shear bands and its relation to the
mechanisms of dilatancy and failure of dense granular soils.” In: Geotechnique 48.4,
pp. 465–481 (cit. on pp. 7, 18).
Oda, Masanobu (1972). “Deformation mechanism of sand in triaxial compression tests.”
In: Soils and Foundations 12.4, pp. 45–63 (cit. on p. 16).
O’Sullivan, Catherine, Jonathan D. Bray, and Shaofan Li (2003). “A new approach for
calculating strain for particulate media.” In: International Journal for Numerical and
Analytical Methods in Geomechanics 27.10, pp. 859–877 (cit. on pp. 28, 59).
Papadopoulos, Lia et al. (2016). “Evolution of network architecture in a granular material
under compression.” In: Physical Review E 94.3, p. 032908 (cit. on p. 1).
Papadopoulos, Lia et al. (2018). “Network analysis of particles and grains.” In: Journal of
Complex Networks 6.4, pp. 485–565 (cit. on pp. 2, 55).
Peters, JF et al. (2005). “Characterization of force chains in granular material.” In: Physical
review E 72.4, p. 041307 (cit. on pp. 16, 17, 59).
Peyneau, Pierre-Emmanuel and Jean-Noël Roux (2008). “Frictionless bead packs have
macroscopic friction, but no dilatancy.” In: Physical review E 78.1, p. 011307 (cit. on
p. 7).
Prévost, J-H and K Höeg (1975). “Soil mechanics and plasticity analysis of strain softening.”
In: Geotechnique 25.2, pp. 279–297 (cit. on p. 8).
Pucilowski, Sebastian and Antoinette Tordesillas (2020). “Rattler wedging and force chain
buckling: metastable attractor dynamics of local grain rearrangements underlie globally
bistable shear banding regime.” In: Granular Matter 22.1, pp. 1–7 (cit. on pp. 55–57).
Bibliography 143
Radjai, Farhang et al. (1996). “Force distributions in dense two-dimensional granular sys-
tems.” In: Physical review letters 77.2, p. 274 (cit. on pp. 16, 59).
Radjai, Farhang et al. (1998). “Bimodal character of stress transmission in granular pack-
ings.” In: Physical review letters 80.1, p. 61 (cit. on p. 18).
Read, H_E and GA Hegemier (1984). “Strain softening of rock, soil and concrete—A review
article.” In: Mechanics of Materials 3.4, pp. 271–294 (cit. on pp. 8, 9).
Rechenmacher, A, S Abedi, and O Chupin (2010). “Evolution of force chains in shear bands
in sands.” In: Géotechnique 60.5, pp. 343–351 (cit. on pp. 57, 63).
Reynolds, Osborne (1885). “LVII. On the dilatancy of media composed of rigid particles
in contact. With experimental illustrations.” In: The London, Edinburgh, and Dublin
Philosophical Magazine and Journal of Science 20.127, pp. 469–481 (cit. on p. 6).
Rocks, Jason W, Andrea J Liu, and Eleni Katifori (2021). “Hidden topological structure of
flow network functionality.” In: Physical Review Letters 126.2, p. 028102 (cit. on pp. 2,
56).
Roscoe, KH, AN Schofield, and CP Wroth (1958). “On the yielding of soils.” In: Géotech-
nique 8.1, pp. 22–53 (cit. on pp. 2, 10, 56).
Rothenburg, L and Nicolaas P Kruyt (2004). “Critical state and evolution of coordina-
tion number in simulated granular materials.” In: International Journal of Solids and
Structures 41.21, pp. 5763–5774 (cit. on pp. 10, 56).
Rowe, Peter W (1962). “The stress-dilatancy relation for static equilibrium of an assem-
bly of particles in contact.” In: Proceedings of the Royal Society of London. Series A.
Mathematical and Physical Sciences 269.1339, pp. 500–527 (cit. on pp. 6, 7).
Sadd, Martin H, Gautam Adhikari, and Francisco Cardoso (2000). “DEM simulation of
wave propagation in granular materials.” In: Powder Technology 109.1-3, pp. 222–233
(cit. on p. 93).
Saitoh, Kuniyasu, Vanessa Magnanimo, and Stefan Luding (2015). “A master equation for
the probability distribution functions of forces in soft particle packings.” In: Soft matter
11.7, pp. 1253–1258 (cit. on p. 16).
Salvatore, Erminio et al. (2017). “Determination of the critical state of granular materials
with triaxial tests.” In: Soils and Foundations 57.5, pp. 733–744 (cit. on p. 28).
Satake, M et al. (1982). “Fabric tensor in granular materials.” In: (cit. on p. 16).
Satake, Masao (1992). “A discrete-mechanical approach to granular materials.” In: Inter-
national journal of engineering science 30.10, pp. 1525–1533 (cit. on pp. 18, 41).
Schall, Peter, David A Weitz, and Frans Spaepen (2007). “Structural rearrangements that
govern flow in colloidal glasses.” In: Science 318.5858, pp. 1895–1899 (cit. on p. 79).
Schofield, Andrew and Peter Wroth (1968). Critical State Soil Mechanics. Vol. 310. McGraw-
Hill London (cit. on pp. 7, 10).
Shahin, Ghassan et al. (2016). “A study of the influence of REV variability in double-scale
FEM× DEM analysis.” In: International Journal for Numerical Methods in Engineering
107.10, pp. 882–900 (cit. on p. 20).
144 Bibliography
Shi, Jingshan and Peijun Guo (2018). “Induced fabric anisotropy of granular materials in
biaxial tests along imposed strain paths.” In: Soils and Foundations 58.2, pp. 249–263
(cit. on p. 34).
Sibille, Luc et al. (2015). “Granular plasticity, a contribution from discrete mechanics.” In:
Journal of the Mechanics and Physics of Solids 75, pp. 119–139 (cit. on p. 31).
Smilauer, V et al. (2015). Yade documentation 2nd ed. (Cit. on pp. 13, 25).
Sterpi, Donatella (1999). “An analysis of geotechnical problems involving strain softening
effects.” In: International Journal for Numerical and Analytical Methods in Geome-
chanics 23.13, pp. 1427–1454 (cit. on p. 9).
Sun, Xulai et al. (2021). “Experimental evidence of detailed balance in granular systems.”
In: arXiv preprint arXiv:2105.01355 (cit. on pp. 2, 57, 63).
Taylor, Donald W (1948). Fundamentals of soil mechanics. Vol. 66. 2. LWW (cit. on p. 7).
Theocharis, Alexandros I et al. (2019). “Necessary and sufficient conditions for reaching
and maintaining critical state.” In: International Journal for Numerical and Analytical
Methods in Geomechanics 43.12, pp. 2041–2055 (cit. on pp. 10, 11, 25, 28, 39, 50).
Tordesillas, A (2007). “Force chain buckling, unjamming transitions and shear banding in
dense granular assemblies.” In: Philosophical Magazine 87.32, pp. 4987–5016 (cit. on
pp. 9, 16, 18, 55, 81).
Tordesillas, Antoinette, Jingyu Shi, and Timothy Tshaikiwsky (2011). “Stress–dilatancy
and force chain evolution.” In: International Journal for Numerical and Analytical
Methods in Geomechanics 35.2, pp. 264–292 (cit. on pp. 7, 8, 18).
Tordesillas, Antoinette, David M Walker, and Qun Lin (2010). “Force cycles and force
chains.” In: Physical Review E 81.1, p. 011302 (cit. on pp. 16, 19, 20, 56).
Tordesillas, Antoinette, Jie Zhang, and Robert Behringer (2009). “Buckling force chains in
dense granular assemblies: physical and numerical experiments.” In: Geomechanics and
Geoengineering: An International Journal 4.1, pp. 3–16 (cit. on pp. 16, 18).
Tordesillas, Antoinette et al. (2012). “Transition dynamics and magic-number-like behavior
of frictional granular clusters.” In: Physical Review E 86.1, p. 011306 (cit. on pp. 9, 55–
57, 63).
Tordesillas, Antoinette et al. (2021). “Spatiotemporal slope stability analytics for failure
estimation (SSSAFE): linking radar data to the fundamental dynamics of granular
failure.” In: Scientific Reports 11.1, pp. 1–18 (cit. on p. 56).
Verdugo, Ramon and Kenji Ishihara (1996). “The steady state of sandy soils.” In: Soils and
Foundations 36.2, pp. 81–91 (cit. on pp. 8, 38).
Veylon, Guillaume (2017). “Modélisation numérique du mécanisme de liquéfaction des sols:
application aux ouvrages hydrauliques.” PhD thesis. Université Grenoble Alpes (cit. on
p. 104).
Walker, David M et al. (2011). “Complex networks in confined comminution.” In: Physical
Review E 84.2, p. 021301 (cit. on p. 56).
Bibliography 145
Walker, David M et al. (2014). “A complex systems analysis of stick-slip dynamics of a

laboratory fault.” In: Chaos: An Interdisciplinary Journal of Nonlinear Science 24.1,
p. 013132 (cit. on p. 55).
Walker, David M et al. (2015). “Self-assembly in a near-frictionless granular material:
conformational structures and transitions in uniaxial cyclic compression of hydrogel
spheres.” In: Soft matter 11.11, pp. 2157–2173 (cit. on pp. 55, 56).
Wan, Richard, Francois Nicot, and Felix Darve (2017). Failure in Geomaterials: a Con-
temporary Treatise. Elsevier (cit. on p. 27).
Wan, Richard G and Pei J Guo (2004). “Stress dilatancy and fabric dependencies on sand
behavior.” In: Journal of Engineering Mechanics 130.6, pp. 635–645 (cit. on pp. 6, 8,
34).
Wang, Rui et al. (2020). “Fabric evolution and dilatancy within anisotropic critical state
theory guided and validated by DEM.” In: International Journal of Solids and Struc-
tures 188, pp. 210–222 (cit. on p. 56).
Wautier, A et al. (2021). “Multiscale modelling of granular materials in boundary value
problems accounting for mesoscale mechanisms.” In: Computers and Geotechnics 134,
p. 104143 (cit. on pp. 20, 93, 97, 99, 129).
Wautier, Antoine, Stéphane Bonelli, and François Nicot (2018a). “Flow impact on granular
force chains and induced instability.” In: Physical Review E 98.4, p. 042909 (cit. on
pp. 16, 57).
Wautier, Antoine, Stephane Bonelli, and François Nicot (2018b). “Micro-inertia origin of
instabilities in granular materials.” In: International Journal for Numerical and Ana-
lytical Methods in Geomechanics 42.9, pp. 1037–1056 (cit. on pp. 75, 125).
Wautier, Antoine, Stéphane Bonelli, and François Nicot (2019a). “DEM investigations of
internal erosion: Grain transport in the light of micromechanics.” In: International
Journal for Numerical and Analytical Methods in Geomechanics 43.1, pp. 339–352 (cit.
on p. 25).
— (2019b). “Rattlers’ contribution to granular plasticity and mechanical stability.” In:
International Journal of Plasticity 112, pp. 172–193 (cit. on p. 57).
Wood, D Muir and K Belkheir (1994). “Strain softening and state parameter for sand
modelling.” In: Géotechnique 44.2, pp. 335–339 (cit. on p. 8).
Wroth, CP and RH Bassett (1965). “A stress–strain relationship for the shearing behaviour
of a sand.” In: Geotechnique 15.1, pp. 32–56 (cit. on p. 7).
Xiong, H, Frédéric Nicot, and ZY Yin (2017). “A three-dimensional micromechanically
based model.” In: International Journal for Numerical and Analytical Methods in Ge-
omechanics 41.17, pp. 1669–1686 (cit. on p. 97).
Xiong, Hao (2017). “Multiscale modeling of granular materials in application to geotech-
nical engineering problems.” PhD thesis. Université Grenoble Alpes (cit. on pp. 113,
129).
146 Bibliographie
Xiong, Hao, Zhen-Yu Yin, and François Nicot (2019). “A multiscale work-analysis approach
for geotechnical structures.” In: International Journal for Numerical and Analytical
Methods in Geomechanics 43.6, pp. 1230–1250 (cit. on pp. 20, 129).
Xiong, Hao et al. (2021). “A novel multi-scale large deformation approach for modelling of
granular collapse.” In: Acta Geotechnica, pp. 1–18 (cit. on pp. 20, 129).
Yan, Beichuan and Richard A Regueiro (2019). “Definition and symmetry of averaged
stress tensor in granular media and its 3D DEM inspection under static and dynamic
conditions.” In: International Journal of Solids and Structures 161, pp. 243–266 (cit. on
p. 99).
Yao, Y-P, Wei Hou, and A-N Zhou (2009). “UH model: three-dimensional unified hardening
model for overconsolidated clays.” In: Geotechnique 59.5, pp. 451–469 (cit. on pp. 8,
10).
Yao, Yang-Ping et al. (2019). “Unified hardening (UH) model for clays and sands.” In:
Computers and Geotechnics 110, pp. 326–343 (cit. on p. 8).
Yao, YP, Lin Liu, and Ting Luo (2016). “UH model for sands.” In: Chinese Journal of
Geotechnical Engineering 38.12, pp. 2147–2153 (cit. on p. 8).
Yin, Zhen-Yu and Ching S Chang (2013). “Stress–dilatancy behavior for sand under load-
ing and unloading conditions.” In: International Journal for Numerical and Analytical
Methods in Geomechanics 37.8, pp. 855–870 (cit. on p. 38).
Zhu, Huaxiang (2015). “Including a meso-structure in multi-scale modeling of granular
soils.” PhD thesis. Université Grenoble Alpes (cit. on p. 113).
Zhu, Huaxiang, François Nicot, and Félix Darve (2016). “Meso-structure organization in
two-dimensional granular materials along biaxial loading path.” In: International Jour-
nal of Solids and Structures 96, pp. 25–37 (cit. on pp. 19, 113).
Zhu, Huaxiang et al. (2016). “On a common critical state in localized and diffuse failure
modes.” In: Journal of the Mechanics and Physics of Solids 95, pp. 112–131 (cit. on
pp. 9, 10, 19, 24, 26, 28, 29, 38, 41, 52, 56, 57, 59, 61, 69, 82).

DENG 2022 Archivage

Uploaded by

Copyright:

Available Formats

DENG 2022 Archivage

Uploaded by

Document Information

Original Description:

Original Title

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

DENG 2022 Archivage

Uploaded by

Copyright:

Available Formats

THÈSE

Pour obtenir le grade de

préparée au sein du Laboratoire INRAE ETNA

Micromécanique de l'état critique et son

used when we created them.”

I would like to extend my sincere gratitude to my co-supervisors Dr. Yannick Thiery

Thanks should also go to Dr. Antoinette Tordesillas for broadening my understanding

I am thankful to French Geological Survey (BRGM) and China Scholarship Council

Multiscale modeling of granular materials has received an increasing attention in recent

Based on these extensive investigations on the critical state, mesostructural transfor-

Keywords: Granular materials, Critical state, Proportional loading path, Dynamical

Mots clés : Matériaux granulaires, État critique, Chemin de chargement proportion-

1.1 Motivation and objectives . . . . . . . . . . . . . . . . . . . . . . . . . . . 1

1.2 Structure of presentation . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3

2 A brief literature review 5

2.1 Shearing behavior of granular materials . . . . . . . . . . . . . . . . . . . . 5

2.1.1 Dilatancy and contractancy . . . . . . . . . . . . . . . . . . . . . . 6

2.1.2 Softening and hardening . . . . . . . . . . . . . . . . . . . . . . . . 8

2.1.3 Critical State . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 10

2.2 Discrete element view of granular materials . . . . . . . . . . . . . . . . . . 11

2.2.1 Discrete Element Method . . . . . . . . . . . . . . . . . . . . . . . 11

2.3 Micromechanical analysis in granular materials . . . . . . . . . . . . . . . . 15

2.3.2 Force chains . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 16

2.3.3 Grain loops . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 18

2.4 Multiscale modeling of granular materials . . . . . . . . . . . . . . . . . . . 20

3 On the attraction power of critical state 23

3.1 DEM simulation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 24

3.2 Mechanical response . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 27

3.2.1 Mechanical response along dilatant proportional strain loading paths 27

3.2.2 Mechanical response along biaxial loading paths . . . . . . . . . . . 28

3.3 p-q-e space analysis . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 30

3.3.1 p-q plane . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 30

3.3.2 p-e plane . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 38

3.4 Fabric-related critical state locus . . . . . . . . . . . . . . . . . . . . . . . 39

3.4.1 Fabric tensor analysis . . . . . . . . . . . . . . . . . . . . . . . . . . 39

3.4.2 Grain loop evolution . . . . . . . . . . . . . . . . . . . . . . . . . . 41

3.5 Mixed proportional strain and biaxial loading paths . . . . . . . . . . . . . 46

3.6 Conclusion and outlook . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 52

4 Dynamical view of the critical state 55

4.1 Numerical setup . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 58

4.2 The so-called critical state . . . . . . . . . . . . . . . . . . . . . . . . . . . 58

4.2.1 Stress-strain analysis . . . . . . . . . . . . . . . . . . . . . . . . . . 58

4.2.2 Force chains and deviatoric stress . . . . . . . . . . . . . . . . . . . 59

4.2.3 Grain loops and volumetric strain . . . . . . . . . . . . . . . . . . . 61

4.3 Hidden dynamics at critical state . . . . . . . . . . . . . . . . . . . . . . . 63

4.3.1 Generating and vanishing process of chained particles . . . . . . . . 63

4.3.2 Generating and vanishing process of grain loops . . . . . . . . . . . 69

4.4 Microstructure reorganization dynamics under different confining pressures 69

4.5 Microstructure dynamics under evolving conditions: memory effects in gran-

4.5.1 Fading process of the initial memory . . . . . . . . . . . . . . . . . 82

4.5.2 Memory fading process along proportional strain test . . . . . . . . 87

5 Critical state and the H-model 95

5.1 Review and analysis of the H-model . . . . . . . . . . . . . . . . . . . . . . 97

5.1.1 The H-model in brief . . . . . . . . . . . . . . . . . . . . . . . . . . 97

5.1.2 Biaxial test at the material point scale . . . . . . . . . . . . . . . . 102

5.1.3 Mesoscale inspection of the H-model during biaxial loading . . . . . 105

5.2 Emergence of critical state in the H-model . . . . . . . . . . . . . . . . . . 114

5.2.1 Method . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 114