PHD Thesis

Statistical geometric analysis of
hard-disk and hard-sphere

microstructures
A Thesis submitted
for the degree of
Doctor of Philosophy
in the Faculty of Engineering
by
V. Senthil Kumar
Department of Chemical Engineering
Indian Institute of Science
Bangalore 560 012 (India)
February 2006
To my teachers
Mallika,
Kanchana, and
Jayashree.
Acknowledgments
I express my sincere gratitude to Prof. V. Kumaran, for allowing me to work
in statistical geometry. His constant guidance and constructive criticism has lead
to a continuous improvement in the quality of the work. His course on statistical
mechanics helped me nourish a deep interest towards the subject. His passion for
research always remains an example for his graduate students to emulate.
I express my sincere gratitude to Prof. Binny Cherayil for a systematic introduc-
tion to the thermodynamic text by Callen.
I thank Prof. Jayant Modak and Prof. Kesava Rao for their understanding and
support on various occasions. I thank Prof. Ganapathy Ayappa for his concern and
support. I thank Prof. Prabhu Nott for allowing me to be the tutor for his thermo-
dynamics course. I thank Prof. Sanjeev Kumar Gupta and Prof. Giridhar for their
counsel.
I thank Dr. B. G. Prakash, Dr. Arun Kumar and Dr. K. Raghunathan for facili-
tating a leave during the last stages of the thesis.
I thank Hrishi, Prashant and Ganesh for linux rescues. I thank Girija, Narayanappa,
Prabhakar and Venkataiah for their eagerness to help.
I thank my lab mates, starting with Sunthar and Rochish in the distant past, to
Partha and Anki in the present, and also others, for lending a ear to my frustrations
and sharing my joys. I appreciate the patience of Krishnan and Vinit for listening to
my boring narrations on hard-sphere packings.
Without the excellent services provided by the Supercomputer Education and
Research Centre and JRD Tata Memorial Library this research work would have
been impossible.
Abstract
In this thesis we show that different statistical geometric measures are useful in
characterizing the microstructures, locating the structural transitions and estimating
the properties. We use the Voronoi polyhedron, the generalization of the Wigner-
Seitz cells for irregular structures, as the fundamental statistical geometric construct
to analyze the thermodynamic, random and homogeneously sheared inelastic hard-
disk and hard-sphere structures. Following are the salient results:
The Voronoi cell volume distributions for hard-disk and hard-sphere uids have
been studied. The distribution of the Voronoi free-volume v
f
, which is the differ-
ence between the actual cell volume and the minimal cell volume at close packing,
is well described by two-parameter and three-parameter gamma distributions. The
free parameter m in both the gamma models is identied as the regularity factor.
The regularity factor is the ratio of the square of the mean and the variance of the
free-volume distribution and it increases as the cell volume distribution becomes
narrower. For the thermodynamic structures, the regularity factor increases with in-
creasing density and it increases sharply across the freezing transition, in response
to the onset of order. The regularity factor also distinguishes between the dense
thermodynamic structures and dense random or quenched structures. The maxi-
mum information entropy (max-ent) formalism, when applied to the gamma distri-
butions, shows that structures of maximum information entropy have an exponential
distribution of v
f
. Simulations carried out using a swelling algorithm indicate that
the dense random packed states approach the distribution predicted by the max-ent
formalism, though the limiting case could not be realized in simulations due to the
structural inhomogeneities introduced by the dense random packing algorithm. Us-
ing the gamma representations of the cell volume distribution, we check the numer-
ical validity of the Cohen-Grest expression [Phys. Rev. B 20, 1077 (1979)] for the
cellular (free-volume) entropy, which is a part of the congurational entropy. The
expression is exact for hard-rod system, and a correction factor equal to the dimen-
sion of the system, D, is found necessary for the hard-disk and hard-sphere systems.
ii
Thus, for the hard-disk and hard-sphere systems, the present analysis establishes a
relationship between the precisely dened Voronoi free-volume (information) en-
tropy and the thermodynamic entropy.
Voronoi tessellation gives a clear denition of geometric neighbors. It is shown
that the n
th
neighbor coordination number (C
n
) and the n
th
neighbor distance dis-
tribution (f
n
(r)) together comprise the total information in the radial distribution
function. The distribution of the number of Voronoi faces (P
n
) is also of interest,
because C
1
is its mean. We have analyzed the C
n
and P
n
for the thermodynamic
and random structures, for the entire density range of hard-disks and hard-spheres.
These statistics are sensitive indicators of microstructure, and they distinguish the
dense thermodynamic and random structures. On freezing, the hard-disk C
n
de-
crease sharply to the hexagonal lattice values. A sharp rise in the hexagon popu-
lation marks the onset of hard-disk freezing transition. In hard-disk random struc-
tures the pentagon and heptagon populations are signicant. In dense hard-disk
structures, both thermodynamic and random, the pentagon and heptagon popula-
tions are nearly identical. Troadec et al. [Euro. Phys. Lett. 42, 167 (1998)]
proved that, due to topological instability a slightly perturbed face-centered cubic
(fcc) lattice of hard-spheres has Voronoi polyhedra with faces 12 to 18, with the
mean at 14. Hence, on freezing transition the hard-sphere C
1
is close to 14 rather
than 12. We demonstrate that this result is thermodynamically consistent. In hard-
sphere random structures, the dodecahedron population decreases with increasing
density. To demonstrate the utility of the neighbor analysis, we estimate the effec-
tive hard-sphere diameter of the Lennard-Jones uid from its C
1
. The estimates are
within 2% deviation from the theoretical results of Barker-Henderson and Weeks-
Chandler-Andersen.
We report the neighbor analysis for the homogeneously sheared inelastic hard-
particle structures, the simplest model for rapid granular matter. The pair distri-
bution function is partitioned into the nth neighbor coordination number (C
n
), and
the nth neighbor position distribution [f
n
(r)]. The distribution of the number of
Voronoi faces (P
n
) is also considered since C
1
is its mean. We report the C
n
and
P
n
for the homogeneously sheared inelastic hard-disk and hard-sphere structures.
These statistics are sensitive to shear ordering transition, the non-equilibrium ana-
iii
logue of the freezing transition. In the near-elastic limit, the sheared uid statistics
approach that of the thermodynamic uid. On shear ordering, due to the onset of
order, the sheared structure C
n
drop to the thermodynamic solid phase value. The
suppression of nucleation by the homogeneous shear is evident in these statistics.
As inelasticity increases, the shear ordering packing fraction increases. In shear or-
dered inelastic hard-sphere structures there is a high incidence of 14-faceted poly-
hedra and a consequent depletion of polyhedra with faces 12, 13, 15 to 18, due
to the formation of body-centered tetragonal (bct) structures. These bct structures
leave a signature like the body-centered cubic structure in the C
n
and P
n
data. On
shear ordering, close-packed layers slide past each other. However, with a velocity-
dependent coefcient of restitution, at a critical shear rate these layers get disordered
or amorphized. We nd that the critical shear rate for amorphization is inversely
proportional to the particle diameter, as compared to the inverse square scaling ob-
served in dense colloidal suspensions.
We report the bond-orientational analysis results for the thermodynamic, ran-
dom and homogeneously sheared inelastic hard-disk and hard-sphere structures.
The thermodynamic structures show a sharp rise in the order across the freezing
transition. The random structures show the absence of crystallization. However, in
random hard-disk congurations local crystallization is high due to the lack of ge-
ometric frustration in two-dimensions. Due to the suppression of crystal nucleation
by the homogeneous shear, the sheared structures get ordered at a packing frac-
tion higher than the thermodynamic freezing packing fraction. On shear ordering,
strings of close-packed hard-disks in two-dimensions and close-packed planes of
hard-spheres in three-dimensions, oriented along the velocity direction, slide past
each other. Such a ow creates a considerable amount of four-fold order in two-
dimensions and bct structure in three-dimensions. These transitions are the ow
analogues of the martensitic transformations occurring in metals due to the stresses
induced by a rapid quench. While the transition in the metallic systems are due to
diffusionless lattice distortions occurring at rates comparable to the speed of sound,
the transition in the ow systems are due to the sliding of close-packed layers past
each other occurring over the ow time scales. In hard-disk structures, using the
bond-orientational analysis we show the presence of four-fold order. When close-
packed spheres slide past each other, in addition to the bct structure, the hexagonal
iv
close-packed (hcp) structure is formed due to the random stacking faults. In sheared
inelastic hard-sphere structures, even though the global bond-orientational analysis
shows that the system is highly ordered, a third-order rotational invariant analysis
shows that only about 40% of the spheres have fcc order, even in the dense and
near-elastic limits, clearly indicating the coexistence of multiple crystalline orders.
Using the Honeycutt-Andersen pair analysis and an analysis based on the 14-faceted
polyhedra having six quadrilateral and eight hexagonal faces, we show the presence
of bct and hcp signatures in shear ordered inelastic hard-spheres. Thus, our analysis
shows that the dense sheared inelastic hard-spheres have a mixture of fcc, bct and
hcp structures.
Contents
1 Introduction 1
1.1 The Voronoi tessellation . . . . . . . . . . . . . . . . . . . . . . . 1
1.2 Thermodynamic and random congurations . . . . . . . . . . . . . 4
1.3 Homogeneously sheared inelastic hard-particle structures . . . . . . 5
1.4 Models for inelastic collision . . . . . . . . . . . . . . . . . . . . . 7
1.5 Overview of the chapters . . . . . . . . . . . . . . . . . . . . . . . 9
2 Voronoi cell volume distribution and congurational entropy of hard-
spheres 11
2.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 12
2.2 Cell volume distribution . . . . . . . . . . . . . . . . . . . . . . . 14
2.3 Maximum information entropy (max-ent) formalism . . . . . . . . 18
2.4 Simulation results . . . . . . . . . . . . . . . . . . . . . . . . . . . 21
2.5 Cellular or free-volume entropy . . . . . . . . . . . . . . . . . . . 29
2.6 Communal entropy . . . . . . . . . . . . . . . . . . . . . . . . . . 40
2.7 Conclusions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 43
Appendix . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 44
2.A Congurational properties . . . . . . . . . . . . . . . . . . . . . . 44
2.B Error analysis . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 45
3 Voronoi neighbor statistics of hard-disks and hard-spheres 51
3.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 52
3.2 Voronoi partitioning of the radial distribution function . . . . . . . . 54
3.3 Hard-rod results . . . . . . . . . . . . . . . . . . . . . . . . . . . . 56
3.4 Hard-disk results . . . . . . . . . . . . . . . . . . . . . . . . . . . 57
3.5 Hard-sphere results . . . . . . . . . . . . . . . . . . . . . . . . . . 61
v
vi Contents
3.6 Estimation of the effective hard-sphere diameter for Lennard-Jones
uid from C
1
. . . . . . . . . . . . . . . . . . . . . . . . . . . . . 71
3.7 Conclusions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 73
Appendix . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 74
3.A Topological instability of fcc lattice . . . . . . . . . . . . . . . . . 74
4 Voronoi neighbor statistics of sheared inelastic hard-disks and hard-
spheres 77
4.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 78
4.2 Voronoi partitioning of the pair distribution function . . . . . . . . . 81
4.3 Hard-disk results . . . . . . . . . . . . . . . . . . . . . . . . . . . 82
4.4 Hard-sphere results . . . . . . . . . . . . . . . . . . . . . . . . . . 88
4.5 Conclusions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 103
5 Bond-orientational analysis of hard-disk and hard-sphere structures 107
5.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 108
5.2 Hard-disk structures . . . . . . . . . . . . . . . . . . . . . . . . . . 110
5.3 Hard-sphere structures . . . . . . . . . . . . . . . . . . . . . . . . 118
5.4 Conclusions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 135
Appendix . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 136
5.A Orthogonality of the quadrilateral faces of the signature polyhedra . 136
References 139
List of Tables
1.1 Salient properties of hard-rods, hard-disks and hard-spheres. . . . . 4
2.1 Properties of two-parameter gamma (2) and three-parameter gamma
(3) distributions. . . . . . . . . . . . . . . . . . . . . . . . . . . . 14
2.2 Parameters of 2 and 3 models for the cell volume distribution of
a D-dimensional Poisson tessellation. . . . . . . . . . . . . . . . . 15
2.3 Equation-of-state and excess entropy equations . . . . . . . . . . . 39
2.4 Virial coefcients for hard-disk and hard-sphere uids . . . . . . . . 40
3.1 Systemsize/shape dependence and thermodynamic consistency checks
for hard-sphere C
1
. . . . . . . . . . . . . . . . . . . . . . . . . . . 65
3.2 Parameters in the the different effective hard-sphere diameter models 70
3.3 Comparison of effective hard-sphere diameter values predicted from
C
1
with those from the theoretical models . . . . . . . . . . . . . . 72
4.1 System size dependence check for sheared inelastic hard-disk struc-
ture C
2
and C
3
. . . . . . . . . . . . . . . . . . . . . . . . . . . . . 87
4.2 System size dependence check C
1
and C
2
for inelastic hard-spheres
sheared in the preferred orientation . . . . . . . . . . . . . . . . . . 93
4.3 C
1
and C
2
for different structured systems . . . . . . . . . . . . . . 94
4.4 P
n
for different structured systems . . . . . . . . . . . . . . . . . . 97
4.5 Shear amorphization with velocity-dependent COR in the preferred
orientation, at different sphere diameters. . . . . . . . . . . . . . . 104
5.1 System size dependence check for sheared inelastic hard-disk struc-
ture f
6
, s
6
and s
8
. . . . . . . . . . . . . . . . . . . . . . . . . . . 119
5.2 System size dependence check for the fraction of fcc clusters in
sheared inelastic hard-sphere structures . . . . . . . . . . . . . . . 125
5.3 System size dependence for the fraction of bond types in sheared
inelastic hard-disk structures . . . . . . . . . . . . . . . . . . . . . 132
vii
viii List of Tables
5.4 Percentage incidence and the orthogonality of the signature polyhe-
dra in thermodynamic and sheared inelastic hard-sphere structures. . 137
List of Figures
1.1 The Voronoi tessellation of a hard-disk conguration . . . . . . . . 2
1.2 A conguration with the Lees-Edwards boundary conditions and
Voronoi tessellation. . . . . . . . . . . . . . . . . . . . . . . . . . 6
2.1 Hard-rods along a line. . . . . . . . . . . . . . . . . . . . . . . . . 16
2.2 Two-dimensional Poisson tessellation cell area distribution. . . . . . 22
2.3 Cell volume distribution for hard-disk thermodynamic structure. . . 22
2.4 Cell volume distribution for hard-disk swelled random structure. . . 23
2.5 2 model m values for hard-disk structures. . . . . . . . . . . . . . 24
2.6 Trend of hard-disk thermodynamic structure cell volume distributions. 25
2.7 Trend of hard-disk swelled random structure cell volume distributions. 25
2.8 2 model m for hard-disk swelled random structures at different
success rates. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 27
2.9 3 model m for hard-disk structures. . . . . . . . . . . . . . . . . . 27
2.10 3 model for hard-disk structures. . . . . . . . . . . . . . . . . . 28
2.11 Error in 3 model as a function of along constant m trajectories,
for hard-disk swelled random structures. . . . . . . . . . . . . . . . 28
2.12 3D Poisson tessellation cell volume distribution. . . . . . . . . . . . 30
2.13 2 model m for hard-sphere structures. . . . . . . . . . . . . . . . 31
2.14 3 model m for hard-sphere structures. . . . . . . . . . . . . . . . 32
2.15 3 model for hard-sphere structures. . . . . . . . . . . . . . . . . 32
2.16 Hard-disk excess entropy prediction from m-data. . . . . . . . . . . 36
2.17 Hard-sphere excess entropy prediction from m-data. . . . . . . . . . 37
2.18 Mean square-error in the cell area models for hard-disk thermody-
namic structures. . . . . . . . . . . . . . . . . . . . . . . . . . . . 47
2.19 Mean square-error in the cell area models for hard-disk swelled ran-
dom structures. . . . . . . . . . . . . . . . . . . . . . . . . . . . . 47
ix
x List of Figures
2.20 Mean absolute relative-error in the cell area models for hard-disk
thermodynamic structures. . . . . . . . . . . . . . . . . . . . . . . 48
2.21 Mean absolute relative-error in the cell area models for hard-disk
swelled random structures. . . . . . . . . . . . . . . . . . . . . . . 48
2.22 Mean square-error in the cell volume models for hard-sphere ther-
modynamic structures. . . . . . . . . . . . . . . . . . . . . . . . . 49
2.23 Mean square-error in the cell volume models for hard-sphere swelled
random structures. . . . . . . . . . . . . . . . . . . . . . . . . . . . 49
2.24 Mean absolute relative-error in the cell volume models for hard-
sphere thermodynamic structures. . . . . . . . . . . . . . . . . . . 50
2.25 Mean absolute relative-error in the cell volume models for hard-
sphere swelled random structures. . . . . . . . . . . . . . . . . . . 50
3.1 The Voronoi tessellation of a hard-disk conguration, showing the
rst and second nearest neighbors of a central disk. . . . . . . . . . 52
3.2 Voronoi partitioning of hard-disk g(r) . . . . . . . . . . . . . . . . 55
3.3 P
n
for two-dimensional Poisson tessellation. . . . . . . . . . . . . . 57
3.4 C
2
for hard-disk thermodynamic and swelled random structures . . 58
3.5 C
3
for hard-disk thermodynamic and swelled random structures. . . 59
3.6 P
6
for hard-disk thermodynamic and swelled random structures . . . 61
3.7 P
n
for hard-disk thermodynamic structures. . . . . . . . . . . . . . 62
3.8 P
n
for hard-disk swelled random structures. . . . . . . . . . . . . . 62
3.9 C
1
for hard-sphere thermodynamic and swelled random structures. . 63
3.10 C
2
for hard-sphere thermodynamic and swelled random structures. . 63
3.11 g
n
(r) for thermodynamic hard-sphere structures. . . . . . . . . . . 66
3.12 P
n
for hard-sphere thermodynamic structures. . . . . . . . . . . . . 67
3.13 P
n
for hard-sphere thermodynamic structures, continued. . . . . . . 67
3.14 P
n
for hard-sphere swelled random structures. . . . . . . . . . . . . 68
3.15 P
n
for hard-sphere swelled random structures, continued . . . . . . 68
3.16 Comparison of P
n
for near regular close packing and near dense
random packing. . . . . . . . . . . . . . . . . . . . . . . . . . . . 69
3.17 Rhombic dodecahedron, the Voronoi cell for the perfect fcc lattice. . 75
3.18 The octahedron formed by the spheres sharing a type-B vertex. . . . 76
3.19 A Voronoi cell in a perturbed fcc lattice. . . . . . . . . . . . . . . . 76
4.1 C
2
for sheared inelastic hard-disk structures. . . . . . . . . . . . . . 85
4.2 C
3
List of Figures xi
4.3 Angular-averaged pair-distribution of sheared structures compared
with the thermodynamic g(r). . . . . . . . . . . . . . . . . . . . . 86
4.4 Pair distribution function for sheared inelastic hard-disks. . . . . . . 86
4.5 P
6
for sheared inelastic hard-disk structures compared with thermo-
dynamic structures. . . . . . . . . . . . . . . . . . . . . . . . . . . 89
4.6 P
n
4.7 C
1
for inelastic hard-spheres sheared in the preferred orientation
compared with thermodynamic structures. . . . . . . . . . . . . . . 91
4.8 C
2
for inelastic hard-spheres sheared in the preferred orientation
compared with thermodynamic structures. . . . . . . . . . . . . . . 92
4.9 Projected pair-distribution function for the preferred orientation shear
of inelastic hard-spheres. . . . . . . . . . . . . . . . . . . . . . . . 92
4.10 g() for the preferred orientation shear of inelastic hard-spheres
compared with the thermodynamic uid and solid phases. . . . . . . 96
4.11 P
n
for sheared inelastic hard-sphere structures. . . . . . . . . . . . 96
4.12 P
n
for sheared inelastic hard-sphere structures, continued. . . . . . 97
4.13 C
1
for sheared inelastic hard-sphere structures at the preferred and
an unpreferred orientation, compared with thermodynamic structures. 98
4.14 Projected pair-distribution function for an unpreferred orientation
shear of inelastic hard-spheres before and at the metastable amor-
phization. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 101
4.15 P
n
for inelastic hard-spheres sheared at an unpreferred orientation,compared
with the random structures. . . . . . . . . . . . . . . . . . . . . . . 101
4.16 Projected pair-distribution function for the preferred orientation shear
before and at the steady state amorphization. . . . . . . . . . . . . . 103
5.1 [
6
[ for hard-disk structures. . . . . . . . . . . . . . . . . . . . . . 111
5.2 [
8
[ for hard-disk structures. . . . . . . . . . . . . . . . . . . . . . 113
5.3 Evolution of f
6
and f
8
in a sheared inelastic hard-disk simulation run.114
5.4 s
6
for thermodynamic and random hard-disk structures. . . . . . . . 115
5.5 s
8
for thermodynamic and random hard-disk structures. . . . . . . . 116
5.6 s
6
for thermodynamic and sheared inelastic hard-disk structures. . . 117
5.7 s
8
for thermodynamic and sheared inelastic hard-disk structures. . . 117
5.8 Q
6
for hard-sphere structures. . . . . . . . . . . . . . . . . . . . . . 120
5.9 Distribution of Q
446
for thermodynamic hard-sphere structures . . . 121
5.10 Distribution of Q
446
for sheared inelastic hard-sphere structures. . . 122
5.11 Fraction of fcc clusters in hard-sphere structures. . . . . . . . . . . 123
5.12 An ideal bct cluster. . . . . . . . . . . . . . . . . . . . . . . . . . . 124
xii List of Figures
5.13 A real bct cluster formed during the homogeneous shear of inelastic
hard-spheres. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 127
5.14 Some of the bonds observed in dense particle systems. . . . . . . . 128
5.15 Fraction of different bond types in thermodynamic hard-sphere struc-
tures. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 129
5.16 Fraction of different bond types in swelled randomhard-sphere struc-
tures. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 129
5.17 Fraction of bond type is in sheared inelastic hard-sphere structures. . 131
5.18 The joint distribution of the lattice ratios for fcc hard-sphere solid
structures. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 132
5.19 The joint distribution of the lattice ratios for sheared inelastic hard-
sphere structures. . . . . . . . . . . . . . . . . . . . . . . . . . . . 133
5.20 The joint distribution of the lattice ratios hcp hard-sphere solid struc-
tures. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 133
Chapter 1
Introduction
In this thesis we show that different statistical geometric measures are useful in
characterizing the microstructures, locating the structural transitions and estimating
the properties. We use the Voronoi polyhedron, the generalization of the Wigner-
Seitz cells for irregular structures, as the fundamental statistical geometric construct
to analyze the thermodynamic, random and homogeneously sheared inelastic hard-
disk and hard-sphere structures.
1.1 The Voronoi tessellation
The Voronoi polyhedron of a point nucleus in space is the innermost polyhedron
formed by the perpendicularly bisecting planes between the given nucleus and all
the other nuclei, Voronoi (1908). The Voronoi tessellation divides a region into
space-lling, non-overlapping convex polyhedra. The salient properties of Voronoi
tessellation are :
Any point inside a Voronoi cell is closer to its nucleus than any other nuclei,
Fig. 1.1. These cells are space-lling and hence a precise denition of local
volume, Rapaport (1983).
It gives a denition of geometric neighbors. The nuclei sharing a common
Voronoi surface are geometric neighbors. Points on the shared surface are
1
2 Introduction
Figure 1.1: The Voronoi tessellation of a hard-disk con guration, with periodic boundary
conditions (PBC). The central box shown in dashed lines.
equidistant to the corresponding pair of nuclei. Hence geometric neighbors
are competing centers in a growth scenario.
Voronoi cells of hard-spheres are irregular at lower packing fractions, but
become regular as the regular close packing is approached. Thus, they are
useful is characterizing all structures, from random to regular.
These properties qualify Voronoi tessellation as an important tool in the structural
analysis of random media like glass, packings, foams, cellular solids, proteins etc.,
see for example Aurenhammer (1991); Okabe et al. (1992); Schliecker (2002).
Voronoi tessellation occurs naturally in growth processes like crystallization and
plant cell growth, Boots (1982). It is used (and was rediscovered under different
names) in various elds like meteorology, geology, ecology, metallography, arche-
ology etc. The statistical distributions of many Voronoi cell properties are reported,
see Zhu et al. (2001), Oger et al. (1996), and references therein. We have used
the algorithm presented in the program listing of Allen and Tildesley (1992) for the
computation of the Voronoi polyhedra. The other geometric computations are from
Chapter 1 3
Preparata et al. (1993) and ORourke (1994).
The Voronoi cells of hard-rods, hard-disks and hard-spheres are segments, poly-
gons and polyhedra respectively, and volume correspondingly means, length, area
and solid volume. Let v
p
be the volume of the hard-particle. Let v be any individual
Voronoi cell volume, and v the average Voronoi cell volume (identical to specic
volume, since the Voronoi cells are space lling). Then, the packing fraction is
given by = v
p
/v. At low the Voronoi cells are irregular, and as increases the
cells become more regular. At the regular close packing, all the cells are identical.
Let v
c
be the Voronoi cell volume at the regular close packing. For hard rods v
c
is
the length of the hard rod, for hard-disks it is the regular hexagon circumscribing
the hard-disk and for hard-spheres it is the rhombic dodecahedron circumscribing
the hard-sphere (corresponding to face-centered cubic (fcc) structure). The packing
fraction at regular close packing is
c
= v
p
/v
c
, and the normalized packing fraction
is y = /
c
.
The packing fractions of physical relevance among the thermodynamic and
metastable hard-structures are:
The freezing (
F
), melting (
M
) packing fractions
The loose random packing (
LRP
) is dened as the lowest density isotropic
packing that can support an innitesimal external load at the limit of acceler-
ation due to gravity tending to zero, Onoda and Liniger (1990).
The dense randompacking (
DRP
) is the highest density homogeneous isotropic
packing.
All these salient packing fractions and the entropy change per particle due to freez-
ing transition are listed in Table 1.1. There is no freezing transition for hard-rod
system. Also there are no random structures for hard-rods since the regular close
packing is the only load bearing structure.
4 Introduction
Table 1.1: Salient properties of hard-rods, hard-disks and hard-spheres.
hard-rod hard-disk hard-sphere
is length diameter diameter
Volume of the particle, v
p

2
/4
3
/6
Cell volume at regular close packing v
c

3
2
/2
3
/
2
Freezing packing fraction,
F
- 0.691
a
0.494
b
Entropy change per particle on freezing s
F
/k
B
- 0.36
a
1.16
b
Melting packing fraction,
M
- 0.716
a
0.545
b
Loose random packing,
LRP
- 0.772 0.002
c
0.555 0.005
d
Dense random packing,
DRP
- 0.82 0.02
e
0.64 0.02
e
Regular close packing,
c
1 /(2
3) /(3
2)
a
From Alder and Wainwright (1962).
b
From Hoover and Ree (1968).
c
From Hinrichsen et al. (1990).
d
From Onoda and Liniger (1990).
e
From Berryman (1983).
1.2 Thermodynamic and random congurations
We have studied two types of thermodynamic hard-particle congurations: the
canonical Monte-Carlo (NVE-MC) and the isothermal-isobaric Monte-Carlo (NPT-
MC). The NVE-MC congurations are made with 50% success rate, i.e. the am-
plitude of the random trial displacement is adjusted such that 50% of the trials lead
to non-overlapping congurations. The NPT-MC algorithm is from Wood (1967).
The results from NPT congurations are ascribed to the average packing fraction.
We generate the random stuctures by a Monte-Carlo adaptation of the algorithm
by Woodcock (1976): swell the nearest neighbors till they touch each other, give
random trial steps (with say 50% success rate as in NVE-MC) for all the particles
and continue the swelling process till desired is attained. If the success rate is
low, the large trial displacements tend to equilibrate the local structures. However,
if success rate is high, the trial displacements are small and the swelling process
Chapter 1 5
locks the the particles into random structures. At any success rate, congurations of
a desired packing fraction are generated through a relay of lower packing fraction
congurations (we use = 0.01 from the point particle limit) with reinitializ-
ing the random number generator at each halt, to avoid the build up of structural
signatures due to the random number generator algorithm. Using different statis-
tical geometric measures, we show that in the high success rate limit the resultant
structures are maximally random. Also, we observe that below the freezing packing
fraction, the random structures are identical to the thermodynamic structures, while
above the freezing packing fractions they are quite distinct from each other.
1.3 Homogeneously sheared inelastic hard-particle structures
We use the Lees and Edwards (1972) boundary condition, to generate homoge-
neously sheared inelastic hard-particle congurations, Fig. 1.2. The simulation box
dimensions are l
x
, l
y
and l
z
. In our simulation geometry, x, y and z are the veloc-
ity, gradient and vorticity directions, respectively. In two-dimensional simulations,
the z direction is absent. The top and the bottom boxes move with velocities +U
and U respectively, with respect to the central box. When a particle crosses the
top/bottom boundary of the central box with a horizontal velocity v
x
, its image en-
ters through the bottom/top with a horizontal velocity (v
x
)
image
= v
x
U. This
induces shear at the top/bottom boundaries of the central box, which then prop-
agates by collisions into the central box. The post-collisional velocity u
of the
smooth particles of identical mass undergoing a binary collision is given in terms of
the pre-collisional velocities as
u
1
= u
1
1 +
2
(w k) k,
u
2
= u
2
+
1 +
2
(w k) k,
where w = u
1
u
2
, k is the unit vector along the line joining the centers of the
colliding particles 1 and 2, and is the coefcient of normal restitution. The energy
input to the system due to shear is lost within the system by inelastic collisions,
hence a thermostat is not required. In dense sheared systems, above a packing
fraction called the shear ordering packing fraction, the system gets ordered. Shear
6 Introduction
PSfrag replacements
+U
U
Figure 1.2: A con guration with the Lees-Edwards boundary conditions and Voronoi tes-
sellation, with PBC.
ordering transition is the nonequilibrium analogue of the thermodynamic freezing
transition. The shear ordering packing fraction, however, depends on the inelastic-
ity of the collision. When the system is below its shear ordering packing fraction,
a linear velocity prole with a shear rate = U/l
y
is induced in the central box.
Above the shear ordering packing fraction, the close-packed layers slide over each
intermittently and a linear velocity prole does not develop, see for example, Fig. 10
of Stevens and Robbins (1993). Further details on the inelastic hard-particle imple-
mentation of the algorithm are available in Campbell (1989) and Goldhirsch and
Tan (1996). In our simulations the initial state is the hexagonal packing for hard-
disks and fcc packing for hard-spheres. To disintegrate the initial lattice, we allow
100 elastic collisions per particle. Then the collisions are made inelastic, and the
granular temperature is monitored. We found that allowing 5000 inelastic collisions
per particle is sufcient to reach the steady state. We have ensured that the struc-
tural statistics are unaffected by the initial conguration, by comparing the results
for regular and random initial congurations.
Chapter 1 7
1.4 Models for inelastic collision
A real collision is characterized using three coefcients: the normal and tangential
coefcients of restitution and a frictional coefcient, Foerster et al. (1994). For
smooth particles, only the normal coefcient of restitution (COR) , needs to be
considered. It is known long since that depends on the impact velocity (v
im
), Ra-
man (1918). However, a constant COR model is still widely used to acquire insights
into the cooperative phenomena in granular matter, Dahl and Hrenya (2004). For
most part of this thesis, we use a constant COR model, since the resultant sheared
structures are shear rate independent, i.e. any nonzero shear rate produces the same
structure and a statistical geometric analysis could be carried out with the two pa-
rameters: packing fraction and the constant COR . However, when we need
to study the shear rate dependent effects, we use a velocity-dependent COR. The
following theoretical results are available for the velocity dependence of in the
plastic and viscoelastic limits:
When the impact velocities are comparable to the yield velocity (v
Y
), due to
plastic deformation, Bridges et al. (1984),
v
1/4
im
. (1.1)
Johnson (1985) showed that, when v
im
v
Y
,
1.18 (v
im
/v
Y
)
1/4
. (1.2)
This asymptotic result is experimentally observed for Nylon (Labous et al.
(1997)) and steel (Li et al. (2002)). For the two-dimensional plastic deforma-
tion v
1/3
im
, Hayakawa and Kuninaka (2002).
When the impact velocities are much lower than the yield velocity, due to
viscoelastic dissipation, 1 v
1/5
im
, Kuwabara and Kono (1987). Poschel
and Schwager (1998); Schwager and Poschel (1998) have shown that, when
v
im
v
Y
,
= 1
1
v
1/5
im
+
2
v
2/5
im

3
v
3/5
im
+ . . . , (1.3)
where
1
and
2
are positive material-dependent constants.
8 Introduction
The above mentioned are the theoretical results. Now, we take note of the models
used in simulations, in addition to the constant COR model.
There is a step model, Poschel et al. (2003), reminiscent of the square-well
approximation for the Lennard-Jones potential,
=
_
1 if v
im
v
c
;
c
if v
im
> v
c
.
,
where v
c
and
c
are constant cut-off values.
Spahn et al. (1997) use the form
=
A
(A
4
+ v
)
1/4
, (1.4)
where A 0.2 0.4 is a t parameter and v
= v
im
/v
Y
is the dimensionless
impact velocity. This form satises 1 when v
0 and the plastic

deformation power dependence (Eq. 1.1). However, it does not accommodate
Johnsons asymptotic result (Eq. 1.2) and the viscoelastic dissipation power
dependence (Eq. 1.3).
McNamara and Falcon (2005) use the piece-wise model for steel spheres in a
vibrated bed,
=
_
1 (1
Y
) v
1/5
if v
im
< v
Y
;
Y
v
1/4
if v
im
> v
Y
.
,
where
Y
is a t parameter, and reproduce experimental trends. This form
does not accommodate the Johnsons asymptotic result (Eq. 1.2), which is
observed in the steel COR data, see Fig. 7 of Li et al. (2002).
We have modied the form of Eq. 1.4 to accommodate the viscoelastic dissipation
(Eq. 1.3) as
=
A + B exp (C v
1/5
)
[(A + B)
4
+ v
]
1/4
, (1.5)
with A = 1.18 to accommodate Johnsons asymptotic result (Eq. 1.2), and B, C
are t parameter. This form accommodates the theoretical results Eqs. 1.1-1.3, and
seems to t well the data for Nylon in Labous et al. (1997) and that for steel in Li
Chapter 1 9
et al. (2002). When we need to study the shear rate dependent effects, we use the
form Eq. 1.5, with B = 0.12 and C = 0.44, they t well the experimental data for
Nylon spheres reported in Labous et al. (1997), they report v
Y
9 m/s.
1.5 Overview of the chapters
In Chapter 2 we analyze the cell volume distributions of the thermodynamic and
swelled random hard-particle structures. In Chapter 3 we report the neighbor anal-
ysis for thermodynamic and swelled random hard-particle structures. In Chapter 4
we extend the neighbor analysis to homogeneously sheared inelastic hard-particle
structures and compare them with the thermodynamic structures. In Chapter 5
we report the bond-orientational analysis for thermodynamic, swelled random and
sheared inelastic hard-particle structures, with particular focus on the microstruc-
tures formed in dense homogeneously sheared inelastic hard-particle structures.
Chapter 2
Voronoi cell volume distribution and
congurational entropy of
hard-spheres
The Voronoi cell volume distributions for hard-disk and hard-sphere uids have
been studied. The distribution of the Voronoi free-volume v
f
, which is the differ-
ence between the actual cell volume and the minimal cell volume at close pack-
ing, is well described by two-parameter and three-parameter gamma distributions.
The free parameter m in both the gamma models is identied as the regularity fac-
tor.The regularity factor is the ratio of the square of the mean and the variance
of the free-volume distribution and it increases as the cell volume distribution be-
comes narrower. For the thermodynamic structures, the regularity factor increases
with increasing density and it increases sharply across the freezing transition, in
response to the onset of order. The regularity factor also distinguishes between the
dense thermodynamic structures and dense random or quenched structures. The
maximum information entropy (max-ent) formalism, when applied to the gamma
distributions, shows that structures of maximum information entropy have an ex-
ponential distribution of v
f
. Simulations carried out using a swelling algorithm
indicate that the dense random packed states approach the distribution predicted by
the max-ent formalism, though the limiting case could not be realized in simula-
tions due to the structural inhomogeneities introduced by the dense random packing
11
12 Voronoi cell volume distribution and con gurational entropy of hard-spheres
algorithm. Using the gamma representations of the cell volume distribution, we
check the numerical validity of the Cohen-Grest expression [Phys. Rev. B 20, 1077
(1979)] for the cellular (free-volume) entropy, which is a part of the congurational
entropy. The expression is exact for hard-rod system, and a correction factor equal
to the dimension of the system, D, is found necessary for the hard-disk and hard-
sphere systems. Thus, for the hard-disk and hard-sphere systems, the present anal-
ysis establishes a relationship between the precisely dened Voronoi free-volume
(information) entropy and the thermodynamic entropy. Some of the results in this
chapter are reported in Kumar and Kumaran (2005a).
2.1 Introduction
We dene the Voronoi free-volume of any hard-particle as the difference between its
Voronoi volume and the minimal cell volume occurring at the regular close packing.
Since there are varied notions of free-volume, we list below the classication of
free-volumes by Bondi (1954):
1. Empty volume = V V
w
, where V is the observed molar volume of the uid
and V
w
is the van der Waals volume of the uid. V
w
is the volume occu-
pied by a molecule which is impenetrable for the other molecules, at a given
temperature. It is the soft potential generalization of the exclusion sphere
concept used in the hard-sphere systems. It is widely used as the molecular
steric descriptor to correlate physiochemical properties (Lepori and Gianni
(2000)) and biological activity (McGowan and Mellors (1986)). The free-
volume used in the uctuating cell theory (Hoover et al. (1979); Reiss (1992))
is an empty volume.
2. (Thermal) Expansion volume = V V
0
, where V
0
is the molar volume of the
substance in its crystalline state at absolute zero temperature. The Voronoi
free-volume used in this thesis is the microscopic version of expansion vol-
ume. The free-volume used in the Doolittle uidity equation (Doolittle (1951))
is an expansion volume, but with V
0
in a hypothetical state without phase
change. This is the most widely used free-volume in the glassy polymer lit-
erature, Haward (1973). However, the other two free-volumes are also used,
Chapter 2 13
Matsuoka and Hale (1997); Yamamoto and Furukawa (2001). There are also
a few experimental measures of free-volume which do not clearly t into this
classication, Consolati et al. (2001).
3. Fluctuation volume = N
A
v
, where v
is the volume swept by the center

of gravity of the molecule due to thermal motion and N
A
is the Avogadros
constant. This is the notion of free-volume used in the lattice or regular cell
theories, Barker (1963).
Section 2.2 analyzes the two-parameter gamma (2) and the three-parameter
gamma (3) distributions used to represent the free-volume distributions of hard-
disk and hard-sphere systems. Since the Voronoi cells are space lling, the average
cell volume is identical to the specic volume. After imposing this constraint, the
2 and 3 distributions have respectively one and two free parameters. The free
parameter m in both the 2 and 3 models is identied as a structural order pa-
rameter called the regularity factor. In Section 2.3 using a maximum information
entropy (max-ent) formalism, based on the notion that for an ordered state all the
cell volumes are identical, we predict that the free-volume distribution is exponen-
tial for an ideal dense random packed state. In Section 2.4 we present the simulation
results for the thermodynamic and random structures of hard-disk and hard-sphere
systems. For the thermodynamic structures, regularity factor increases with increas-
ing density and it increases sharply across the freezing transition, in response to the
onset of order. The regularity factor also distinguishes between the dense thermo-
dynamic structures and dense random structures. The max-ent prediction for the
dense random packing seems to be approached, but not exactly reached due to the
structural inhomogeneities introduced by the dense random packing algorithm used.
In Section 2.5, using the 2 model, we check the Cohen-Grest ansatz for congura-
tional entropy, and nd that ansatz is exact for hard-rod system, and a factor equal
to the dimension of the system is found missing in the original ansatz, for the hard-
disk and hard-sphere systems. This rst-order homogeneity of the entropy in the
dimension of the system, D, is anticipated if the number of states of an independent
particle increases with D as 2
D
. In Section 2.6, we show that the free-volume the-
ory prescription for the communal entropy is untenable, atleast for the hard-particle
uid states.
Table 2.1: Properties of two-parameter gamma (2) and three-parameter gamma (3) distribu-
tions.
Property 2 distribution 3 distribution
a b
Distribution, f(v
f
)

m
(m)
v
(m1)
f
e
v
f

m
2
(
m
2
)
v
(m/1)
f
e
v
f
0 v
f
< m, > 0 m, , > 0
Mean = v
f
m
m+
Variance =
2
(v
f
) = v
2
f
(v
f
)
2 m(m+1)
2

_
m
_
2
=
m
m+2
m+
_
2
Standard deviation
Mean
=
(v
f
)
v
f
1
m
_
m+2
m+
2
1
_1
2
a
With = 1 and using (m + 1) = m(m), the 3 results reduce to the 2 results.
b
Generally the 3 model is written as f(v
f
) =

m
(
m
)
v
( m1)
f
e
v
f
, see for example Tanemura
(2001, 2003). Using m = m/ we get the above form. This form has the advantage that m is the
regularity factor in both the 2 and 3 models.
2.2 Cell volume distribution
It is customary to t the cell volume distribution data for random points (Poisson-
Voronoi tessellation) to a 2 or a 3 distribution, given in Table 2.1 (read with
v
f
= v). The subscript 0 is used to indicate the low density or the Poisson limit .
Using the specic volume criteria we get,
0
=
_
_
m
0
v
2 model;
_
m
0
+
0
2
0
m
0
2
0
v
_
0
3 model.
(2.1)
The reported best-t values of the free parameters (m
0
for 2 and m
0
,
0
for 3
models) are given in Table 2.2. Note that only the 1D results are exact. Heuristic
arguments in Weaire et al. (1986) show that the cell area distributions for a two-
dimensional (2D) Poisson tessellation can be approximated by a 2 distribution.
For a hard-core tessellation, some studies like Hermann et al. (1989) and Gotoh
Chapter 2 15
Table 2.2: Parameters of 2 and 3 models for the cell volume distribution of a D-dimensional
Poisson tessellation.
D m
0

0
a
Reference
1 2 1 Exact result, Kiang (1966)
2 3.5 1 Kiangs revised value, as in Weaire et al. (1986)
2 3.61 1 Weaire et al. (1986)
2 3.57 1 DiCenzo and Wertheim (1989)
2 3.57782 1 Current thesis, Fig. 2.2
2 3.5700 1.0787 Hinde and Miles (1980)
2 3.57371 1.07805 Tanemura (2001)
2 3.63454 1.09577 Current thesis, Fig. 2.2
3 6 1 Kiang (1966)
3 5.56 1 Andrade and Fortes (1988)
3 5.56219 1 Current thesis, Fig. 2.12
3 5.59434 1.16391 Tanemura (2001)
3 5.68147 1.19361 Current thesis, Fig. 2.12
4 8.41715 1.29553 Tanemura (2001)
a
0
= 1 implies a 2 model and
0
= 1 a 3 model.
PSfrag replacements
l +
v
P
0
P
1
P
2
x
1
x
2
x
1f
x
2f
Figure 2.1: Hard-rods along a line. Hard-rod length is .
(1993) t the 2 model for the cell volume, i.e. v
f
= v in Table 2.1. The volume
of any Voronoi cell will be greater or equal to v
c
, hence the above t gives an un-
physical non-zero value for
_
vc
0
f(v)dv. Alternatively, for a hard-core tessellation,
the 2 or 3 models can be tted with v
f
= v v
c
in Table 2.1, where v
f
is the
Voronoi free-volume. In Section 2.5, we show that this denition of free-volume
gets the thermodynamic singularity correctly at the regular close packing. In Hin-
richsen et al. (1990) the Voronoi free-volume t is used in the study of loose random
packing. Note that this t accommodates the gamma representation at the Poisson
limit, where = 0 and hence v
c
= 0 and v
f
= v. Now we show that the notion of
Voronoi free-volume arises spontaneously in the hard-rod system. For the hard-rod
system the nearest neighbor distance distribution function, f(x), is exactly known,
see Fisher (1964),
f(x) =
_
_
_
0 if x < ;
1
(v)
exp
_
(x)
(v)
_
if x .
(2.2)
In Fig. 2.1, let x
i
be the distance between the centers of the hard-rods P
i1
and
P
i
and let x
if
be the free distance between their tips, then x
if
= x
i
. Then the
Voronoi segment for P
1
is given by v = (x
1
+ x
2
)/2 = (x
1f
+ x
2f
)/2 . Noting
that in 1D v
c
= , we have v
f
= v v
c
= (x
1f
+x
2f
)/2. Say l = 2v
f
= x
1f
+x
2f
.
Here x
1f
and x
2f
independent, since the the free distances x
if
do not percolate.
Then the cumulative distribution function of l is
F(l) =
_
l
0
dx
2f
_
lx
2f
0
dx
1f
f(x
1f
) f(x
2f
). (2.3)
Chapter 2 17
Using f(x
f
) = a exp (ax
f
) with a = 1/(v ) gives
F(l) =
_
l
0
a e
a x
2f
dx
2f
_
lx
2f
0
a e
a x
1f
dx
1f
= 1 exp (a l) a l exp (a l).
Then the probability distribution function for l is given by,
f(l) =
dF(l)
dl
= a
2
l exp (a l).
Using l = 2v
f
in f(v
f
)dv
f
= f(l)dl we get,
f(v
f
) = (2a)
2
v
f
exp (2av
f
).
With = 2a and m = 2, it can be written as,
f(v
f
) =

m
(m)
v
(m1)
f
exp (v
f
).
Thus for the hard-rod system at any , the Voronoi free-volume follows a 2 distri-
bution with m = 2. This suggests tting a 2 or its generalization the 3 distribu-
tion (Table 2.1) for f(v
f
) in hard-disk and hard-sphere systems. The free parameter
in the 2 model is m, and the free parameters in the 3 model are m and . is
computed using the specic volume criteria as
=
_
_
m
v(1y)
2 model;
_
(
m+
2
)
(
m
2
)v(1y)
_
3 model.
(2.4)
Here we have used v
f
= v v
c
= v(1 y). Note that at y = 1, diverges and
variance becomes zero, so that the gamma distribution reduces to the Dirac delta
distribution. The simulation results in Section 2.4 show that m does not diverge at
the regular close packed limit.
In the 2 model, the ratio of the standard deviation and mean is 1/
m (Table
2.1). This shows that as m increases the spread of the distribution about the mean
decreases, i. e. the cells become more regular. Hence the m is called regularity fac-
tor in Gotoh (1993) (This work however used 2 model for f(v) rather than f(v
f
),
as mentioned earlier). As y increases, the thermodynamic structures become more
regular and hence m increases. At the freezing transition, due to the onset of order,
there is a sharp increase in m. Thus m is a useful scalar parameter characterizing
the hard-disk and hard-sphere structures. In the hard-rod system, since m = 2 for
any , the ratio of standard deviation to the mean is constant, a hallmark of the
absence of the uid-solid transition in this system.
A physical interpretation for the parameter m in the 3 model can be obtained
by using the n asymptotic expansion from Spanier and Oldham (1987),
(n + c)
(n)
n
c
_
1 +
c(c 1)
2n
+
c(c 1)(c 2)(3c 1)
24n
2
+
_
, (2.5)
with n = m/
2
and c = 1/, 2/ successively in Table 2.1 gives,
_
Standard deviation
Mean
_
2
=
1
m
+
(1
2
)
2m
2
+ .
When is close to unity or m is large, the series can be truncated from the second
term. Hence,
Standard deviation
Mean

1
m
.
Thus, in the 3 model m is the regularity factor, when is close to unity or m is
large. If both the models are tted to the same set of data, the m of the 3 model
will be nearly equal to the mof the 2 model. This correspondence can be observed
for the Poisson tessellation case in Table 2.2. This correspondence is shown for
hard-disk and hard-sphere systems in Section 2.4. From the regularity factors in
Table 2.2, it can be noted that the Poisson tessellation becomes more regular as the
dimension increases.
2.3 Maximum information entropy (max-ent) formalism
The uncertainty or the information entropy (Shannon (1948); Jaynes (1957)) of a
discrete distribution is s
I
= k
B
P
i
log P
i
, where P
i
is the probability of the
i
th
outcome and the summation is over all the outcomes. The analogous denition
for a continuous distribution f(x) is s
I
= k
B
_
f(x) log [f(x)] dx, where the
integration is over the full domain of x. As an illustration, consider the uncertainty
of a coin toss, s
I
= k
B
[p log p + q log q], where p is the probability of a head
and q = 1 p is the probability of a tail. It can be easily seen that the uncertainty
Chapter 2 19
of the outcome is maximum when p = q = 1/2. This physically means that when
p = q = 1/2 one would need maximum information (the complete knowledge
of the dynamics of a toss and the initial conditions) to predict the outcome. One
could as well arrive at this value of p by the principle of equal apriori probability
(PEAP), which assigns equal probability for all the outcomes when no other infor-
mation is available. However, in a complex situation where PEAP does not apply
or the complete understanding of the system is lacking, the max-ent formalism is
a powerful method to estimate the most probable distribution, Jaynes (1979). For
example, Englman et al. (1987) have derived the size distribution of fragments in
a disintegration process (say, blasting) showing good agreement with experimental
observations, using the max-ent formalism with the energy and volume constraints
on the fragments.
In general, the information entropy need not be related to the thermodynamic
entropy, since the system under consideration itself need not be in equilibrium, as
in the above mentioned disintegration process. However, if the system is in ther-
modynamic equilibrium and if the distribution under consideration contains all the
relevant information (to the desired level of description), then the information en-
tropy will be identical to the thermodynamic entropy. For example, in a monatomic
uid of N particles in a container of constant volume and temperature, the informa-
tion entropy dened on the N-particle position and velocity distribution function,
f
N
(r
1
, . . . , r
N
, v
1
, . . . , v
N
), where r
i
and v
i
are the position and velocity of the
i
th
particle, is identical to the thermodynamic entropy, Jaynes (1965). Now, if we
employ PEAP for the energy states (i.e. states with identical energy have identical
probability), then f
N
(r
1
, . . . , v
N
) = f[E(r
1
, . . . , v
N
)], and f(E) contains all the
information about the system, to the level of macroscopic averaging. Hence, the in-
formation entropy dened on f(E) is also identical to the thermodynamic entropy,
even though the microscopic information content of f(E) is much less than that in
f
N
(r
1
, . . . , v
N
), Jaynes (1957).
Now, the uncertainty or the information entropy dened on the cell volume dis-
tribution is
s
I
= k
B
_

vc
f(v) log [f(v)] dv.
Transforming the independent variable to the free-volume v
f
= v v
c
gives,
s
I
= k
B
_

0
f(v
f
) log [f(v
f
)] dv
f
. (2.6)
This denition of information entropy employs the equality of all the cell volumes
as the measure of order. At the regular close packing, all the cell volumes are
identical and the free-volume distribution is a Dirac-delta distribution and hence s
I
is negative innity, i. e. we need minimum information (the crystals unit cell) to
construct these congurations. Using the method of Lagrange multipliers, with the
mean free-volume condition (v
f
= v v
c
), it can be easily shown that an expo-
nential free-volume distribution has maximum information entropy. From Section
2.2, we know that the Poisson tessellation (ideal gas) cell volume distribution is not
exponential. The Poisson conguration does not maximize s
I
because these cong-
urations could be easily constructed using a uniform random number generator. We
might speculate that the DRP (or the jammed packings in general) might have an
exponential free-volume distribution, since we need maximum information (all the
particle positions) to construct the conguration. The algorithm we have used pro-
duces dense random structures with nearly but not exactly exponential free-volume
distribution, further details in Section 2.4.
The gamma distributions contain the exponential distribution as a limiting case
(m = 1 for 2 model and , m = 1 for 3 model). Hence, it is obvious that even
within these two family of distributions, the exponential distribution maximizes the
information entropy. However, we showthe explicit analysis for the 2 model, since
we require an intermediate result for Section 2.5. The 2 model, on integration
gives,
s
I
k
B
= log [v(1 y)] (m1)(m) + log [(m)] log (m) + m, (2.7)
where () is the digamma function. Here we have used a standard integral from
Gradshteyn and Ryzhik (2000),
_

0
x
m1
e
x
log (x) dx =
(m)
m
[(m) log ()] , (2.8)
and then eliminated using Eq. 2.4. In the 2 model, the free parameter m is a
function of y and the type of structures (thermodynamic or otherwise), hence s
I
is
Chapter 2 21
a functional of m. To maximize s
I
, we set,
m
_
s
I
k
B
_
= (m1)
_
1
m

(1)
(m)
_
= 0, (2.9)
where
(n)
() are the polygamma functions. It can be easily checked that the solu-
tion m = 1 is a maximum, using
(1)
(1) =
2
/6. We will pursue the connection of
s
I
with the thermodynamic entropy in Section 2.5, for use therein we rewrite Eq. 2.7
as s
I
/k
B
= log [v(1 y)] +(m), where (m) = (m1)(m) + log [(m)]
log (m) + m. Setting y = 0, gives the information entropy for the ideal gas at the
same specic volume as s
0
I
/k
B
= log (v) + (m
0
). Then the excess information
entropy is
s
E
I
k
B
= log (1 y) + (m) (m
0
). (2.10)
2.4 Simulation results
The cell volume distribution for 2D Poisson tessellation is given in Fig. 2.2. The 2
model has a positive error at the peak (shown in inset) and a compensating negative
error near the origin. The 3 model gives a better t, however at the cost of an
additional parameter. This statistical superiority of the 3 model over the 2 model
holds for the entire density range, for both hard-disk and hard-sphere structures, as
shown in the appendix 2.B by a mean square error analysis.
Compare the degree of t for the canonical Monte Carlo (NVE-MC) and the
swelled random structures, both at = 0.82 in Figs. 2.3 and 2.4 (after ignoring
the greater scatter in the random structure data due to lesser averaging). A bimodal
f(v) (as in Fig. 2.4) indicates the existence of dense and lean regions, due to the
formation of crystallites. The most appropriate method to quantify the crystallite
concentration in a random hard-disk conguration is to classify the hard-disks as
solid-like or uid-like based on the bond orientational order parameter (Glaser and
Clark (1993)), and compute the fraction of solid-like disks. This analysis for the
swelled random structures is presented in Chapter 5. Alternatively, one could moni-
tor the population of hexagons as a measure of crystallite concentration, as shown in
Chapter 3. Local crystallization, however is typical of random packing algorithms,
especially in the hard-disk system due to the lack of geometric frustration, Hinrich-
0 1 2 3 4
0
0.2
0.4
0.6
0.8
0.5 0.6 0.7 0.8 0.9
0.76
0.78
0.8
0.82
0.84
0.86
PSfrag replacements
v
f
(
v
)
Figure 2.2: 2D Poisson tessellation cell area distribution. Simulation data (), 2-model
() and 3-model (). Averaged over 10
5
con gurations of 1000 points with periodic
boundary conditions (PBC). The best- t parameter values given in Table 2.2. The cell
volume is scaled by the speci c volume.
0.85 0.9 0.95 1 1.05 1.1
0
2
4
6
8
10
12
14
16
18
PSfrag replacements
v
f
(
v
)
Figure 2.3: Cell volume distribution for hard-disk thermodynamic structure at = 0.82.
Simulation data (), 2-model () and 3-model (). Averaged over 10000 con gu-
rations of 256 hard-disks with PBC.
Chapter 2 23
0.9 1 1.1 1.2
0
2
4
6
8
10
PSfrag replacements
v
f
(
v
)
Figure 2.4: Cell volume distribution for hard-disk swelled random structure at = 0.82.
Simulation data (), 2-model ( ) and 3-model (). Averaged over 1000 con gu-
rations of 256 hard-disks with PBC. Note the bimodal nature of the distribution.
sen et al. (1990). For example, the cell volume distribution in Fig.62 of Glaser and
Clark (1993), in addition to the two distinct modes, has a large-volume tail due to
the grain boundaries around the crystallites. Due to such structural inhomogeneities
caused by the DRP algorithm, the gamma distributions give a poor t near
DRP
(see appendix 2.B) and hence the max-ent predictions are approached but not ex-
actly realized in the following results. The good performance of the 3 model for
the homogeneous thermodynamic structures across the entire range (see appendix
2.B) and the approach to the max-ent predictions, makes us speculate that an ideal
DRP algorithm producing homogeneous structures might satisfy the max-ent pre-
dictions.
The 2 model m values for NVE-MC, isothermal-isobaric Monte Carlo (NPT-
MC) and swelled random hard-disk structures are given in Fig. 2.5, from which we
observe the following:
For thermodynamic structures (NVE and NPT) as increases, cells become
0 0.2 0.4 0.6 0.8
0
2
4
6
8
10
12
14
16
PSfrag replacements
m
Figure 2.5: 2 model m values for hard-disk structures. NVE-MC (), NPT-MC (),
swelled random at 50% success rate (), Poisson limit () and max-ent prediction at dense
random packing limit (). Averaged over 10000 con gurations for NVE/NPT and 1000
con gurations for swelled random structures of 256 hard-disks with PBC.
Chapter 2 25
0.8 0.9 1 1.1 1.2 1.3 1.4
0
5
10
15
20
25
30
0.90
0.85
0.80
0.75
0.70
0.65
PSfrag replacements
v
f
(
v
)
Figure 2.6: Trend of hard-disk thermodynamic structure cell volume distributions. For
>
F
, as increases the cell volume distribution becomes narrow and at
c
, it degenerates
to a Dirac delta distribution located at v
c
. All lengths are scaled by the disk diameter.
0.8 1 1.2 1.4 1.6
0
2
4
6
8
10
0.82
0.80
0.78
0.75
0.70
0.65
PSfrag replacements
v
f
(
v
)
Figure 2.7: Trend of hard-disk swelled random structure cell volume distributions. For
>
F
, as increases the ratio of standard deviation to mean (1/
m) of the cell volume

distribution increases. All lengths are scaled by the disk diameter. The max-ent prediction
is shown by the broken line().
more regular and hence m increases. Due to onset of order at freezing transi-
tion, the thermodynamic structures have a sharp increase in m. The thermo-
dynamic structures terminate at the regular close packing.
The NPT and NVE structures are identical, it can be explained as follows:
Since the radial distribution function of an NPT ensemble at its average is
identical to that of an NVE ensemble at the same , the local neighborhoods
being identical, the cell volume distributions are identical. The differences
between NVE and NPT structures seen on the solid branch may be ascribed
to the limited sampling.
For <
F
the swelled random structures become more regular (i.e. m
increases) as increases, and this behavior is identical to that of the thermo-
dynamic uid structures. However, >
F
the swelled random structures
become more irregular (i.e. m decreases) with increase in , and terminate
at the dense random packing limit. This behavior is drastically different from
that of the thermodynamic solid structures, and can be clearly seen by com-
paring Figs. 2.6 and 2.7.
The swelled random structures can be made with different success rates. As
mentioned earlier, at low success rates the large trial displacements tend to equili-
brate the local non-equilibrium structures and hence the m value approaches that of
the thermodynamic structures at the same . This can be observed in Fig. 2.8, for
>
F
swelled random structures with 10% success rate. Fig. 2.8 shows that with
increasing success rates, all the structures lie on a unique branch of dense random
structures. At any success rate, attaining a desired through a relay of lower con-
gurations with small and reinitializing the random number generator at each
halt, avoids the build up of structural signatures due to the random number generator
algorithm. Structures made in a small relay (here = 0.01) are independent
of the success rates. Structures made in a large relay (say > 0.1) depend on
the success rates, and these structures have an m value lesser than that of the ther-
modynamic structures at the same . This method can be used to generate structures
having an m-value lower than the thermodynamic structures. Structures having an
m-value greater than the thermodynamic structures can be made by shrinking the
disk diameter of denser thermodynamic structures.
Chapter 2 27
0 0.2 0.4 0.6 0.8
0
2
4
6
8
10
PSfrag replacements
m
Figure 2.8: 2 model m for hard-disk swelled random structures for different success rates.
Success rate 10% (), 30% (), 50% (), 70% (), 90% (+), Poisson limit () and max-ent
prediction at dense random packing limit (). Averaged over 1000 con gurations of 256
hard-disks with PBC.
0 0.2 0.4 0.6 0.8
0
2
4
6
8
10
12
14
16
PSfrag replacements
m
Figure 2.9: 3 model m for hard-disk structures. NVE-MC (), NPT-MC (), 50% success
rate swelled random structure (), Poisson limit () and max-ent prediction at dense random
packing limit (). Averaging as in Fig. 2.5.
0 0.2 0.4 0.6 0.8
0
0.2
0.4
0.6
0.8
1
1.2
1.4
1.6
PSfrag replacements
Figure 2.10: 3 model for hard-disk structures. NVE-MC (), NPT-MC (), 50% suc-
cess rate swelled random structures (), Poisson limit () and max-ent prediction at dense
random packing limit (). Averaging as in Fig. 2.5.
0.2 0.4 0.6 0.8 1
10
3
10
2
10
1
10
0
PSfrag replacements
M
e
a
n
S
q
u
a
r
e
E
r
r
o
r
Figure 2.11: Error in 3 model as a function of along constant mtrajectories, for hard-disk
swelled random structures at 50% success rate and = 0.55 (, m = 8.4981), = 0.65
( , m = 9.1465), = 0.73 (., m = 8.1272) and = 0.83 ( , m = 2.2827).
The at the minimum error () are the optimized values reported in Fig. 2.10.
Chapter 2 29
The m and values for the 3 model for the free-volume distribution of hard-
disk structures are given in Figs. 2.9 and 2.10. From Figs. 2.5 and 2.9 it is observed
that the 3 model m compares well with 2 model m, within a few percent differ-
ence, when m is large or is close to unity. In Fig. 2.10, it is interesting to note that
the swelled random structures approach the max-ent prediction ( = 1) at the dense
random packing limit with drastic changes. These drastic trend changes are not
an artifact of curve-tting, since we got identical results with different optimizer
algorithms (Nelder-Mead and Gauss-Newton) and different initial estimates. The
optimizer, after the initial excursion, settles into a nearly constant m trajectory (that
which corresponds to the regularity factor) on the error surface (a function of m
and ), and searches for a value which minimizes the error along this trajectory.
The error surface does not have multiple local minima in the working range of the
parameters (m ranging from 1 to 35 and ranging from 0.1 to 2), as seen from
Fig. 2.11. Since the error magnitudes vary widely for the state points considered,
in Fig. 2.11 a logarithmic scale is used for the y-axis. Though some of the curves
appear at in this scale, the minima are distinct in the linear scale.
Now we present the analogous results in three-dimensions (3D). The 3D Pois-
son tessellation cell volume distribution is given in Fig. 2.12. The 2 model m
values for the hard-sphere structures are given in Fig. 2.13. The m and values for
the 3 model for the free-volume distribution of hard-sphere structures are given in
Figs. 2.14 and 2.15. The approach of the swelled random hard-sphere structures to-
wards the max-ent predictions are less satisfactory than that in the hard-disk system,
even though the trend in m (Fig. 2.14) is towards unity, the trend in (Fig. 2.15) is
away from unity as DRP is approached. The parameter m in the 2 and 3 models
are nearly identical inspite of being nearly 2, since the m values are large.
2.5 Cellular or free-volume entropy
Appendix 2.A gives the thermodynamic denition of the congurational properties.
In the context of regular cell theory, Kirkwood (1950) showed that the congura-
tional entropy can be partitioned as s
conf
= s
cell
+s
com
, where s
cell
is the cellular en-
tropy arising due to the single occupancy of the regular cells and s
com
is the commu-
0 0.5 1 1.5 2 2.5 3
0
0.2
0.4
0.6
0.8
1
PSfrag replacements
v
f
(
v
)
Figure 2.12: 3D Poisson tessellation cell volume distribution. Simulation data (), 2-
model () and 3-model (). Averaged over 1000 con gurations of 1000 points with
PBC. The best- t parameter values given in Table 2.2. The cell volume is scaled by the
speci c volume.
nal entropy arising due to the multiple occupancy of the regular cells. In the context
of free-volume theory of glasses and supercooled liquids, Cohen and Grest (1979);
Grest and Cohen (1981); Cohen and Grest (1982) assumed that the above partition-
ing holds as well for the irregular cellular structure dened by the Voronoi polyhedra
of the particles. They introduced the expression s
fv
= k
B
_
f(v) log [f(v)]dv, in
analogy with the entropy of mixing, where f(v) is the Voronoi cell volume distri-
bution. (Here, we use the notation s
fv
instead of s
cell
, since s
fv
corresponds to the
entropy associated with Voronoi free-volume distributions). To validate the ansatz,
we independently compute and compare the left and right hand sides of the ansatz
for hard-disk and hard-sphere systems. The right hand side is computed from the
free-volume distributions discussed in Section 2.4 and the left hand side is com-
puted independently from the thermodynamic data (details below). To the best of
our knowledge, such an analysis is not reported for any system.
Chapter 2 31
0 0.2 0.4 0.6
0
10
20
30
40
50
PSfrag replacements
m
Figure 2.13: The 2-model m for hard-sphere structures. NVE-MC (), swelled random
at 50% success rate (), Poisson limit () and max-ent prediction at dense random packing
limit (). Averaged over 1000 con gurations of 256 hard-spheres with PBC.
0 0.2 0.4 0.6
0
10
20
30
40
50
PSfrag replacements
m
Figure 2.14: 3 model m for hard-sphere structures. NVE-MC (), swelled random at 50%
success rate (), Poisson limit () and max-ent prediction at dense random packing limit
(). Averaging as in Fig. 2.13.
0 0.2 0.4 0.6
0.5
1
1.5
2
PSfrag replacements
Figure 2.15: 3 model for hard-sphere structures. NVE-MC (), swelled random at 50%
success rate (), Poisson limit () and max-ent prediction at dense random packing limit
(). Averaging as in Fig. 2.13.
Chapter 2 33
For the Voronoi free-volume or cellular entropy we take the Cohen-Grest ex-
pression as
s
fv
= k
B
_

vc
f(v) log [f(v)]dv, (2.11)
allowing a density independent proportionality constant (if is density depen-
dent, the ansatz is of negligible value), its necessity will be clear below. Cohen and
Grest (1979) dene the communal entropy, based on the existence of liquid clusters,
as the entropy associated with the accessibility of all of the congurational volume
within the nite liquid clusters and within the innite cluster, when present. For
densities above the freezing density very few uid clusters exist and hence the com-
munal entropy is a negligible fraction of the congurational entropy. In the regular
close packed limit, the communal entropy is zero and hence the congurational en-
tropy is purely cellular. We identify by comparing the cellular entropy with the
congurational entropy near the regular close packing. For ease of comparison, we
express all entropies as excess entropies with the reference ideal gas at the same
specic volume. Now the above condition is expressed as
lim
c
s
E
conf
= s
E
fv
, (2.12)
and the ansatz Eq. 2.11 is rewritten as s
E
fv
= s
E
I
, where the s
E
I
is the excess
information entropy. Using Eq. 2.10 for s
E
I
we get,
s
E
fv
k
B
= log (1 y) + [(m) (m
0
)]. (2.13)
To compute s
E
conf
we use the thermodynamic identity (proof in Appendix 2.A)
s
E
conf
= s
E
v
, where s
E
v
is the excess entropy of a uid with the reference ideal gas at
the same specic volume. For hard-spheres s
E
v
is computed from the EOS as (Hill
(1956)),
s
E
v
k
B
=
_
y
0
Z 1
y
dy, (2.14)
where Z = pv/(k
B
T) is the compressibility factor. Differentiating Eq. 2.14 with
respect to y and rearranging gives
Z = 1 y
d
dy
_
s
E
v
k
B
_
. (2.15)
Using Eqs. 2.12, 2.13, 2.15 and the identity s
E
conf
= s
E
v
gives,
lim
c
Z =
y
1 y
+ 1 y
_
m
__
dm
dy
_
. (2.16)
Hoover (1966) show that the form of the dense hard-sphere solid EOS is
Z =
D
a
+ c
0
+ c
1
a + c
2
a
2
+ , (2.17)
where a = (v v
c
)/v
c
= (1 y)/y is the dimensionless excess free-volume, see
Table 2.3. Comparing Eqs. 2.16 and 2.17, and noting from Figs. 2.5 and 2.13 that
(dm/dy) does not diverge at y = 1, we identify = D. Before proceeding further,
we would like to emphasize the wide implications of this identication.
The low density and high temperature limit of any system of particles with
short ranged interaction, is identical to the low density limit of hard-spheres.
Hence this identication holds in the ideal gas limit of any system of parti-
cles. But then is density independent, hence the identication holds for any
system at any density. This generality is anticipated since depends only on
the dimension of the system.
It should be noted that the information of the particular form used to t the the
free-volume distribution (in this thesis the gamma distribution) is contained
in of Eq. 2.16. And this information is irrelevant for the identication =
D, as long as it does not contribute a divergent term to the right hand side
of Eq. 2.16 at y = 1. Hence this identication is exact, even though the
gamma representations are merely good approximations. However, note that
the denition of the free-volume as v
f
= v v
c
is crucial in recovering the
thermodynamic singularity correctly at the regular close packing. Any other
denition of free-volume like v
f
= v v
p
where v
p
is the particle volume or
v
f
= v v
opt
where v
opt
is some optimized minimum cell volume (used in
Hanson (1983)) will not recover the thermodynamic singularity at the regular
close packing, when used with the Cohen-Grest ansatz.
With the identication = D, the free-volume or cellular entropy is rst-
order homogeneous in D. Note that the last two terms on the right-hand side
Chapter 2 35
of Eq. 2.27, arising from integration of momenta, are also rst-order homo-
geneous in D. This is a general behavior, as shown by the following sim-
ple model. Consider a system of N independent hard-spheres rattling inside
some volume and suppose that, at any instant, the direction of displacement
is the only relevant variable. Then, at any instant, an independent sphere can
displace in x, y or z directions, hence it has 2
D
states, here D = 3.
The number of states available to the system is = (2
D
)
N
= 2
ND
. Hence,
the entropy, S = k
B
log = Nk
B
Dlog 2, is rst-order homogeneous in D.
This behavior is not observed in the Ising model since the number of states is
identically two (up and down) for an independent spin in any dimension. But
systems in which the number of states of an independent particle increases
with D as 2
D
, have an entropy rst-order homogeneous in D.
For static powders, the communal entropy is identically zero. Hence, the
ensemble averaged entropy for similarly prepared powders is purely cellu-
lar. Hence, with the identication = D in Eq. 2.11, the entropy of static
powders is fully characterized by the free-volume distributions. Even if the
gamma representations are a poor t for the free-volume distributions, a nu-
merical integration of the right hand side of Eq. 2.11 could be effected.
Now we show the validity of the identication = D, for each dimension
separately. In the hard-rod system there is no solid-uid transition. Hence, a hard-
rod conguration at any is a dilute solid, and not a uid. Since there are no uid
clusters in the hard-rod system, its communal entropy (in the Cohen-Grest sense)
is identically zero and the congurational entropy is purely cellular. From Section
2.2, for hard-rods we have m = m
0
= 2 and hence dm/dy = 0. Using these with
= 1 in Eqs. 2.16 and 2.13 we recover the exact results: the Tonks EOS (Tonks
(1936)) Z = (1 y)
1
and s
E
v
/k
B
= log (1 y). Thus, for the hard-rod system,
Eq. 2.11 is exact with = 1.
For the hard-disk and hard-sphere systems, we compute excess entropy s
E
conf
(=
s
E
v
) form the different EOS reported in the literature, using Eq. 2.14. These expres-
sion are given in Table 2.3. The virial coefcients used in Table 2.3 are given in
Table 2.4. After freezing transition, the congurational entropy should be slightly
greater than the free-volume or cellular entropy, since the communal entropy is a
0 0.2 0.4 0.6 0.8
15
10
5
0
0.55 0.6 0.65 0.7 0.75 0.8
6
5.5
5
4.5
4
3.5
3
2.5
2
PSfrag replacements
s
E v
/
k
B
Figure 2.16: Hard-disk excess entropy prediction from m-data. Voronoi-NVE (), Voronoi-
50% swelled random structure (), virial EOS ( ), uid MD t ( ), regular cell
theory ( ) and composite EOS (). The Voronoi prediction is got by using the m-data
from Fig. 2.5 in Eq. 2.11 with = 2. The equations for the other data are given in Table
2.3.
Chapter 2 37
0 0.2 0.4 0.6
15
10
5
0
PSfrag replacements
s
E v
/
k
B
Figure 2.17: Hard-sphere excess entropy prediction from m-data. Voronoi-NVE (),
Voronoi-50% swelled random structure (), virial EOS ( ), uid MD t ( ), reg-
ular cell theory ( ) and composite EOS (). The Voronoi prediction is got by using the
m-data from Fig. 2.13 in Eq. 2.11 with = 3. The equations for the other data are given in
Table 2.3.
small non-negative contribution. The equality holds in the limit of regular close
packing, where the communal entropy vanishes. Using the Voronoi m-data for the
hard-disk and hard-sphere systems from Figs. 2.5 and 2.13, on the right hand side
of Eq. 2.11 with = 2 and 3 respectively, we get the Voronoi estimates of the
free-volume or cellular entropy. These results are plotted in Figs. 2.16 and 2.17.
From these gures, we see that the free-volume or cellular entropy matches with
the congurational entropy for the thermodynamic solid phase, except for the small
non-negative communal entropy (see Fig. 2.16 inset). Note that, in these gures,
the vertical distance between the composite EOS curve and the NVE-MC Voronoi
data is the excess communal entropy. For hard-sphere composite EOS equation
in Table 2.3, we have taken the entropy change during freezing, s
F
/k
B
as 0.92
(instead of the approximate value 1.16 given in Table 1.1), so that the excess com-
munal entropy is non-negative. This revision is justied as follows: By compar-
ing the entropy difference between an ideal gas (with the congurational integral
Q
0
conf
= V
N
/N! = v
N
N
N
/N!) and a regular cellular solid at the same specic
volume (Q
conf
= v
N
), it can be easily seen that s
F
/k
B
for smooth elastic hard
hyper-spheres cannot be greater than unity.
C
h
a
p
t
e
r
2
3
9
Table 2.3: EOS and Excess Entropy equations plotted in Figs. 2.16 and 2.17.
System Z,
s
E
v
k
B
and notes Reference for EOS
HD/HS Virial Expansion Z = 1 + B
2
+ B
3

2
+ B
4

3
+ van Rensburg (1993)
s
E
v
k
B
=
_
B
2
+
B
3
2

2
+
B
4
3

3
+
_
B
2
to B
8
in Table 2.4
HD MD uid data t Z =
1+A +B
2
(1)
2
Maeso and Solana (1993)
s
E
v
k
B
=
(1+A+B)
(1)
(B 1) log (1 )
A = 0.05833, B = 0.01267
HD MD solid data t Z = 2/a + 1.90 + 0.67a + 1.5a
2
Young and Alder (1979)
where a =
vvc
vc
=
1
y
1
HS MD uid data t Z =
1++
2
3
(1)
3
Carnahan and Starling (1970)
or the Carnahan-Starling EOS
s
E
v
k
B
=
(43)
(1)
2
HS MD solid data t Z = 3/a + 2.566 + 0.55a 1.19a
2
+ 5.95a
3
Young and Alder (1979)
a as above
HD/HS composite EOS
s
E
v
k
=
_

0
Z
f
1
d, for <
F
Z
f
is uid t given above
s
E
v
k
B
=
_

F
0
Z
f
1
d
s
F
k
B
F
Zs1
d, for >
F
Z
s
is solid t given above
s
F
/k
B
= 0.36 for HD
s
F
/k
B
= 0.92 for HS
HD/HS Cell theory Z =
_
1 y
1
D
_
1
Wood (1952)
s
E
v
k
B
= D log (1 y
1
D
)
D =dimension.
Table 2.4: Virial coef cients for hard-disk and hard-sphere uids, from Sanchez (1994)
n HD B
n
HS B
n
2 2 4
3 3.128 017 8 10
4 4.257 854 18.364 77
5 5.336 897 28.224 5
6 6.362 6 39.74
7 7.351 53.54
8 8.338 70.8
2.6 Communal entropy
Consider an ensemble of similarly prepared N-particle congurations, thermody-
namic or metastable. The free-volume theory Cohen and Grest (1979); Grest and
Cohen (1981); Cohen and Grest (1982) assumes that there exists a critical local vol-
ume v
cr
, such that any cell can be classied as uid-like if v > v
cr
, and solid-like
if v < v
cr
. The classication of a particle as solid-like implies local jamming and
not local crystalline order. In a N-particle conguration, say there are N
F
uid-like
cells and (N N
F
) solid-like cells. The fraction of uid-like cells is p = N
F
)/N,
where ) indicates an ensemble average. Or equivalently,
p =
_

vcr
f(v) dv. (2.18)
For a particle with Voronoi volume v, the volume above the critical value is assumed
to be available for diffusive movements, hence called the diffusive free-volume v
d
,
i.e.
v
d
=
_
0 for v < v
cr
;
v v
cr
for v > v
cr
.
(2.19)
The mean diffusive free-volume is
v
d
=
_
vcr
(v v
cr
)f(v) dv
_
vcr
f(v) dv
. (2.20)
Chapter 2 41
If two uid-like cells are geometric neighbors they are said to be interconnected.
While some uid-like cells are isolated, some uid-like cells are interconnected.
A uid-like cluster is a set of interconnected uid-like cells. An isolated uid-
like cell is accounted as a uid-like cluster of unit size. A uid conguration (i.e.
at densities below the uid-solid transition) has a spanning or an innite (when
periodic boundary conditions are considered) uid-like cluster. In a conguration
having N
F
uid-like cells, say N
form the spanning cluster and (N

F
N
) form
nite size uid-like clusters. Percolation probability, P
perc
= N
)/N
F
), is the
probability that a uid-like cell is within the spanning cluster. The size distribution
of the nite uid-like clusters, P
n
, is normalized such that
n=1
+
n P
n
=
N
F
) N
)
N
F
)
, (2.21)
where 1
+
is the set of the nite size clusters 1, 2, 3, . . .. For a given choice
of v
cr
, the state of the congurations is fully characterized (under the uid-like /
solid-like classication) by the two descriptors p and P
n
, since the other quantities
are computed as shown below:
N
F
) = Np,
N
F
) N
) = Np
n=1
+
n P
n
, and
P
perc
= 1
n=1
+
nP
n
.
Free volume theory assumes that the exchange of free-volume between the uid-
like cells in a uid-like cluster and its neighboring solid-like cells is very slow com-
pared to the exchange of free-volume between the uid-like cells within the same
uid-like cluster. Or equivalently, the uid-like cells are assumed to exchange their
free-volume among the other uid-like cells in the same uid-like cluster, without
any restriction from its neighboring solid-like cells. However, for a diffusive jump
to occur within a uid-like cluster of size n, the total diffusive free-volume of the
cluster should be greater than the particle volume v
p
, i.e.
n
i=1
(v
i
v
cr
) > v
p
. (2.22)
A uid-like cluster satisfying this criteria is called a uid cluster, since a diffusive
jump can occur within this cluster. The minimum size of a uid cluster is n
f
=
v
p
/v
d
, where v
d
is the average diffusive free-volume, given by Eq. 2.20.
Cohen and Grest (1982) have shown that the contribution to communal entropy
by a uid cluster of size n (with n n
f
) is nk
B
. The number of uid clusters of
size n is NpP
n
. Hence, the communal entropy contribution due to all nite uid
clusters is
Nk
B
p
n=n
+
f
nP
n
,
where n
+
f
is the set of nite size clusters n
f
, (n
f
+1), (n
f
+2), . . .. The number
of particles in the innite cluster is N
) = NpP
perc
and the communal entropy
contribution due to the innite cluster is Nk
B
pP
perc
. Hence, the total communal
entropy due to the nite uid clusters and innite cluster is
S
com
= Nk
B
p
_
_
_
n=n
+
f
nP
n
+ P
perc
_
_
_
= Nk
B
p
_
1
n
f
1
n=1
nP
n
_
. (2.23)
In the ideal gas limit, all the cells are uid-like and all the cells form the innite
uid-like cluster, i.e. N
) = N
F
) = N, hence p = 1. Also P
n
is undened since
there are no nite uid-like clusters. Hence, S
0
com
= Nk
B
. The excess communal
entropy with respect to ideal gas at same specic volume is
S
E
com
= Nk
B
p
_
1
n
f
1
n=1
nP
n
_
Nk
B
.
Thus, the excess communal entropy per particle
s
E
com
k
B
= p
_
1
n
f
1
n=1
nP
n
_
1. (2.24)
Thus, the communal entropy is specied by the two descriptors p and P
n
, and v
cr
alone remains unspecied. From Eq. 2.24, it is easily seen that s
E
com
/k
B
is negative,
because p < 1 and
_
1
n
f
1
n=1
nP
n
_
< 1.
Chapter 2 43
However, from Fig. 2.17 it is seen that s
E
com
/k
B
, which is the vertical distance from
the excess free-volume entropy for thermodynamic structures to the excess con-
gurational entropy for the composite EOS data, is positive. This shows that the
free-volume theory prescription for the communal entropy is untenable, atleast for
hard-particle uid states. At the ideal gas limit i.e. = 0, s
E
com
/k
B
is zero, be-
cause the reference state is ideal gas itself. For densities above the freezing tran-
sition i.e. >
F
, s
E
com
/k
B
is zero because we subtract the entropy change due
to freezing (which is approximately equal to the communal entropy of ideal gas).
In the range 0 < <
F
, s
E
com
/k
B
increases with increasing at low densities
and then decreases after a maximum. Note that a similar trend is observed in the
hard-sphere diffusivity data due to collective effects, Alder and Wainwright (1967);
Speedy (1987).
2.7 Conclusions
The Voronoi free-volume distributions of the hard-disk and hard-sphere systems are
well represented by a two-parameter (2) or a three-parameter gamma (3) distri-
bution, Table 2.1. Since the Voronoi cells are space-lling, the average cell volume
is identical to the specic volume. After imposing this constraint constraint, m is
the free parameter in the 2 model and m, are the free parameters in the 3 model.
It is shown that the parameter min the 2 and the 3 models is the regularity factor.
For the thermodynamic structures, regularity factor increases with increasing den-
sity and it increases sharply across the freezing transition, in response to the onset
of order. The regularity factor also distinguishes between the dense thermodynamic
structures and dense random structures. The maximum information entropy formal-
ism predicts that the free-volume distribution at the dense random packed state is
an exponential distribution. This prediction seems to be approached, but not ex-
actly reached due to the structural inhomogeneities induced by the dense random
packing algorithm. This suggests a precise denition for dense random packing
as the packing at which the information entropy of the Voronoi free-volume dis-
tribution is maximized. The connection between the free-volume entropy and the
thermodynamic entropy was examined in the limit of regular close packing where
the communal part of the entropy vanishes. We nd that the Cohen-Grest expres-
sion relating the thermodynamic and the free-volume entropy is exact for hard-rod
system, and a correction factor equal to the dimension of the system is necessary
for hard-disk and hard-sphere systems. This correction factor is anticipated if the
number of states of an independent particle increases with the dimension of the
system, D, as 2
D
. We have shown that the free-volume theory prescription for the
communal entropy is untenable, atleast for the hard-particle uid states .
Appendix
2.A Congurational properties
The canonical partition function for N indistinguishable classical particles of mass
m in D dimensions is
Q =
_
2mk
B
T
h
2
_D N
2
Q
conf
,
where Q
conf
is the congurational integral,
Q
conf
=
1
N!
_
V

_
V
exp ( U) dr
1
dr
N
,
and U(r
1
, . . . , r
N
) is the potential energy of the systemin the conguration r
1
, . . . , r
N
.
The Helmholtz free energy is F = k
B
T log Q and the entropy is
S =
_
F
T
_
N,V
= k
B
log Q + k
B
T
_
log Q
T
_
N,V
. (2.25)
The congurational analogues (Levelt and Cohen (1964)) are dened as F
conf
=
k
B
T log Q
conf
and
S
conf
=
_
(F
conf
)
T
_
N,V
= k
B
log (Q
conf
) + k
B
T
_
log (Q
conf
)
T
_
N,V
. (2.26)
From the Eqs. 2.25 and 2.26,we get
S = S
conf
+
D
2
Nk
B
log
_
2mk
B
T
h
2
_
+
D
2
Nk
B
(2.27)
Chapter 2 45
Replacing S and S
conf
respectively with S
0
and S
0
conf
we get the entropy of ideal
gas. Then the excess entropy S
E
v
= S
E
conf
. Note that the reference ideal gas has same
specic volume as the uid under consideration, due to the constancy of N and V
in the canonical ensemble. For ideal gas with Q
0
conf
= V
N
/N! and the Stirlings
approximation, log N! = N log N N, we have S
0
conf
= Nk
B
+ Nk
B
log v.
2.B Error analysis
The performance of different models tted to the same data set are assessed using
the mean square-error dened as,
mean square-error =
1
M
M
i=1
_
f
i
f
i
_
2
, (2.28)
where M is the number of data bins, i is the bin index, f
i
is the free-volume dis-
tribution data and

f
i
is the model prediction at v = v
i
. The mean square-error in
t for hard-disk NVE and swelled random structures are given in Figs. 2.18 and
2.19. These gures show that the 3 model is statistically better than the 2 model
over the entire range. Note that the mean square-error dened in Eq. 2.28, can be
used to compare the performance of different models tted to the same data set, but
cannot be used to assess the performance of a model over different data sets. For
example, as density increases the free-volume distribution get narrower and taller,
hence will have a larger mean square-error than a shorter distribution at a lower den-
sity, even though both have the same degree of t (in a visual sense). This implies
that error plots like Fig. 2.18 should be read just vertically, and the trend across
different data sets has no relevance. In Fig. 2.18, the mean square-error near the
regular close packing shoots to a large value since even a small mist (in the vi-
sual sense) in a tall and narrow distribution accumulates a large value to the mean
square-error.
To assess the performance of a model for different data sets we use the mean
absolute relative-error, dened as
mean absolute relative-error =
1
M
M
i=1
f
i
f
i
f
i
. (2.29)
During nonlinear regression, the mean square-error is taken as the objective func-
tion for minimization and not the mean absolute relative-error, because the former
measure favors a close t for the large-f i.e. the signicant features rather than
the small-f features, while the latter measure typically acquires large values due to
small errors at the small-f features. The mean absolute relative-errors for hard-disk
NVE and swelled random structures are given in Figs. 2.20 and 2.21 respectively.
Fig. 2.20 shows that the performance of the 3 model as tends to
c
is nearly sim-
ilar to that at the ideal gas limit (see Fig. 2.4). However, a misreading of Fig. 2.18
can lead to the wrong conclusion that the error increases as tends to
c
. Fig. 2.21
shows that the performance of the 3 model for swelled random structures steadily
deteriorates for
F
. Similar trends are observed in the analogous error plots for
the hard-sphere system, Figs. 2.22 to 2.25.
Chapter 2 47
0 0.2 0.4 0.6 0.8
10
10
10
5
10
0
10
5
10
10
PSfrag replacements
m
e
a
n
s
q
.
e
r
r
o
r
Figure 2.18: Mean square-error in the cell area models for hard-disk NVE thermodynamic
structures. 2-model () and 3-model (). Averaging as in Fig. 2.5.
0 0.2 0.4 0.6 0.8
10
10
10
5
10
0
PSfrag replacements
m
e
a
n
s
q
.
e
r
r
o
r
Figure 2.19: Mean square-error in the cell area models for hard-disk swelled (50% success
rate) random structures. 2-model () and 3-model (). Averaging as in Fig. 2.5.
0 0.2 0.4 0.6 0.8
10
1
10
0
PSfrag replacements
m
e
a
n
a
b
s
.
r
e
l
.
e
r
r
o
r
Figure 2.20: Mean absolute relative-error in the cell area models for hard-disk NVE ther-
modynamic structures. 2-model () and 3-model (). Averaging as in Fig. 2.5.
0 0.2 0.4 0.6 0.8
10
1
10
0
PSfrag replacements
m
e
a
n
a
b
s
.
r
e
l
.
e
r
r
o
r
Figure 2.21: Mean absolute relative-error in the cell area models for hard-disk swelled (50%
success rate) random structures. 2-model () and 3-model (). Averaging as in Fig. 2.5.
Chapter 2 49
0 0.2 0.4 0.6
10
10
10
5
10
0
PSfrag replacements
m
e
a
n
s
q
.
e
r
r
o
r
Figure 2.22: Mean square-error in the cell volume models for hard-sphere NVE thermody-
namic structures. 2-model () and 3-model (). Averaging as in Fig. 2.13.
0 0.1 0.2 0.3 0.4 0.5 0.6
10
10
10
5
10
0
PSfrag replacements
m
e
a
n
s
q
.
e
r
r
o
r
Figure 2.23: Mean square-error in the cell volume models for hard-sphere swelled (50%
success rate) random structures. 2-model () and 3-model (). Averaging as in Fig. 2.13.
0 0.2 0.4 0.6
10
1
10
0
PSfrag replacements
m
e
a
n
a
b
s
.
r
e
l
.
e
r
r
o
r
Figure 2.24: Mean absolute relative-error in the cell volume models for hard-sphere NVE
thermodynamic structures. 2-model () and 3-model (). Averaging as in Fig. 2.13.
0 0.1 0.2 0.3 0.4 0.5 0.6
10
1
10
0
PSfrag replacements
m
e
a
n
a
b
s
.
r
e
l
.
e
r
r
o
r
Figure 2.25: Mean absolute relative-error in the cell volume models for hard-sphere swelled
(50% success rate) random structures. 2-model () and 3-model (). Averaging as in
Fig. 2.13.
Chapter 3
Voronoi neighbor statistics of
hard-disks and hard-spheres
Voronoi tessellation gives a clear denition of geometric neighbors. It is shown
that the nth neighbor coordination number (C
n
) and the nth neighbor distance dis-
tribution (f
n
(r)) together comprise the total information in the radial distribution
function. The distribution of the number of Voronoi faces (P
n
) is also of inter-
est, because C
1
is its mean. We have analyzed the C
n
and P
n
for the thermody-
namic and swelled random structures, for the entire density range of hard-disks
and hard-spheres. These statistics are sensitive indicators of microstructure, and
they distinguish thermodynamic and swelled random structures. On freezing, the
hard-disk C
n
decrease sharply to the hexagonal lattice values. A sharp rise in the
hexagon population marks the onset of hard-disk freezing transition. In hard-disk
random structures the pentagon and heptagon populations are signicant. In dense
hard-disk structures, both thermodynamic and random, the pentagon and heptagon
populations seem identical. Troadec et al. (1998) proved that, due to topological in-
stability a slightly perturbed face-centered cubic lattice of hard-spheres has Voronoi
polyhedra with faces 12 to 18, with the mean at 14. Hence, on freezing transition
the hard-sphere C
1
is close to 14 rather than 12. We demonstrate that this result
is thermodynamically consistent. In hard-sphere random structures, the dodecahe-
dron population decreases with increasing density. To demonstrate the utility of the
neighbor analysis, we estimate the effective hard-sphere diameter of the Lennard-
51
52 Voronoi neighbor statistics of hard-disks and hard-spheres
Figure 3.1: The Voronoi tessellation of a hard-disk con guration, with periodic boundary
conditions (PBC). The central box shown in dashed lines. The rst and second neighbors
of a central disk () are shown linked in dashed lines.
Jones uid from its C
1
. The estimates are within 2% deviation from the theoreti-
cal results of Barker-Henderson, Weeks-Chandler-Andersen and its modication by
Lado. These results are reported in Kumar and Kumaran (2005b)
3.1 Introduction
In this chapter we report the Voronoi neighbor analysis for thermodynamic and
swelled random hard-disk and hard-sphere structures. Section 3.2 introduces the
Voronoi neighbor statistics. Let a central spheres geometric neighbors be called
rst neighbors, i. e. the rst layer of neighbors. The rst neighbors neighbors
(which are themselves not rst neighbors) are the second neighbors, and so on,
Fig. 3.1. Thus, all the spheres surrounding a central sphere can be partitioned layer-
wise, and characterized by nth neighbor coordination number (C
n
) and nth neighbor
Chapter 3 53
distance distribution function (f
n
(r)). In Section 3.2 it is shown that the total in-
formation contained in the radial distribution function, g(r), can be partitioned into
these sets of neighbor statistics. The distribution of the number of Voronoi bounding
surfaces (P
n
) is also of interest, because C
1
is its mean. These neighbor statistics are
sensitive microstructural indicators, and they distinguish the dense thermodynamic
and random structures. For hard rod system these neighbor statistics are exactly
known, given in Section 3.3. For hard-disk and hard-sphere systems, we review the
neighbor statistics reported in the literature, and report C
n
and P
n
for the canonical
Monte Carlo (NVE-MC) and swelled random structures in Section 3.4 and 3.5.
The low density swelled random structures are identical to the thermodynamic
structures. The dense swelled random structures are quite distinct from the thermo-
dynamic structures, and are presumed to terminate at the dense random packing.
However, as shown by Torquato et al. (2000), by negotiating disorder with order we
are able to generate congurations denser than the dense random packing. The C
1
for random hard-sphere structures produced by this algorithm matches reasonably
with the dense random packing experimental results of Finney (1970).
On freezing, the hard-disk C
n
(for n > 1) decrease sharply to the hexagonal
lattice values. A sharp rise in the hexagon population marks the onset of hard-
disk freezing transition. In dense hard-disk random structures the pentagon and
heptagon populations are signicant. In dense hard-disk structures, both thermody-
namic and random, the pentagon and heptagon populations nearly identical. Due to
topological instability, proved by Troadec et al. (1998) (outlined in Appendix 3.A),
a slightly perturbed face-centered cubic (fcc) lattice of hard-spheres has Voronoi
polyhedra with faces 12 to 18, with the mean at 14. Hence, on freezing transition
the hard-sphere C
1
is close to 14 rather than 12. We demonstrate that this result is
consistent with thermodynamic data. In hard-sphere random structures, the dodec-
ahedron population decreases with increasing density.
The notion of effective hard-sphere diameter (EHSD) for dense soft potential
uids has a long history, beginning with Boltzmann himself, see the recent review
Silva et al. (1998). It is known, both through simulations and experiments, that the
structure of a dense soft potential uid is nearly identical to that of the hard-sphere
uid having a particular diameter. This diameter is the EHSD. In Section 3.6, we
show that using the equality of C
1
of the Lennard-Jones uid to the hard-sphere
uid at some packing fraction, it is possible to estimate the effective hard-sphere
diameter to within 2% deviation from the theoretical results of Barker-Henderson,
Weeks-Chandler-Andersen and its modication by Lado.
3.2 Voronoi partitioning of the radial distribution function
Let N
n
be the number of nth neighbors around a central sphere, then the nth neigh-
bor coordination number is C
n
= N
n
), where ) denotes the ensemble average.
The radial distribution function is computed as
g(r) =
(r)
=
1
N
r
)
V
r
, (3.1)
where V
r
is the volume of the shell between r and r +dr around the central sphere
and N
r
is the number of spheres with their centers in the shell between r and r+dr.
The spheres around the central sphere can be partitioned layer-wise as
N
r
=
n=1
N
n
r
, (3.2)
where N
n
r
is the number of nth neighbors with their centers in the shell between r
and r + dr. Using Eq. 3.2 in Eq. 3.1, g(r) can be partitioned as
g(r) =
1
n=1
N
n
r
)
V
r
=
n=1
g
n
(r), (3.3)
where g
n
(r) is the nth neighbor radial distribution function. Fig. 3.2 illustrates a
Voronoi partitioning for hard-disk g(r). Such a partitioning was rst reported by
Rahman (1966), and recently by Lavrik and Voloshin (2001).
The nth neighbor distance distribution function, f
n
(r), is dened such that
f
n
(r) dr is the fraction of the nth neighbors at a distance r to r + dr, then,
f
n
(r) dr =
N
n
r
)
_
0
N
n
r
dr
_ =
N
n
r
)
C
n
. (3.4)
Chapter 3 55
1 2 3 4 5 6 7
0
0.5
1
1.5
2
2.5
3
PSfrag replacements
r
g
(
r
)
g
1 g
2
g
3 g
4
g(r)
g
sum
Figure 3.2: Voronoi partitioning of hard-disk thermodynamic structure g(r), = 0.50,
g
sum
= g
1
+ g
2
+ g
3
+ g
4
. The g
n
(r) and g
sum
are shown in thick lines and g(r) in thin
line.
Here, we have used
_
0
N
n
r
dr
_
= N
n
) = C
n
. Using Eq. 3.3 and Eq. 3.4, we
get,
g
n
(r) =
C
n
f
n
(r)
Vr
dr
=
C
n
f
n
(r)
S
r
, (3.5)
where S
r
is the spherical surface area at r, S
r
= 2 in one-dimension (1D), S
r
= 2r
in two-dimensions (2D) and S
r
= 4r
2
in three-dimensions (3D). Using Eq. 3.5 in
Eq. 3.3, we get,
g(r) =
1
S
r
n=1
C
n
f
n
(r). (3.6)
The Eq. 3.6 shows that the two Voronoi neighbor statistics, C
n
() and f
n
(r; )
together contain the thermodynamic information in g(r; ). We consider another
Voronoi statistic, the distribution of the number of bounding surfaces of the Voronoi
cell, P
n
. It is identical to the distribution of the number of the rst neighbors, hence
C
1
=
n P
n
. We will show in Section 3.4 and 3.5, that these neighbor statis-
tics are sensitive indicators of microstructure which distinguish thermodynamic and
swelled random structures.
In hard-sphere systems the compressibility factor, Z = p/(k
B
T), is related to
the radial distribution function at contact g() as
Z = 1 + B
2
g(), (3.7)
where B
2
is the second virial coefcient, B
2
= for hard rods, B
2
= (/2)
2
for
hard-disks and B
2
= (2/3)
3
for hard-spheres. Now g() = g
1
(), since a sphere
in contact is necessarily a rst neighbor. Eq. 3.5 gives g
1
() = C
1
f
1
()/( S
).
Note that B
2
/S
= /(2D), where D is the dimensionality of the system. Using

these in Eq. 3.7 gives
Z = 1 +

2D
C
1
f
1
(). (3.8)
Thus, for the hard-sphere systems, the two neighbor statistic values C
1
and f
1
()
contain the thermodynamic information.
For any nondegenerate
1
2Dtessellation with periodic boundary conditions (PBC)
or with a large number of particles, C
1
= 6 exactly, Meijering (1953); Collins
(1968). Using this in Eq. 3.8, for hard-disks we have Z = 1 + (3/2)f
1
(). This
result was derived by Ogawa and Tanemura (1974) using a different, but less gen-
eral method, while the above derivation is valid for any dimensions and shows the
role of C
1
.
3.3 Hard-rod results
For hard rod system g
n
(r) is exactly known (Fisher (1964)),
g
n
(r) =
_
0 if r < n ;
v (rn)
n1
(n1)! (v)
n
exp
_
rn
vn
_
if r n .
(3.9)
Here, S
r
= 2 and C
n
= 2. Using this result in Eq. 3.5 gives f
n
(r) = (1/v) g
n
(r).
Using this with Eq. 3.9 gives f
1
() = 1/(v ). Using these results in Eq. 3.8
1
A 2D Voronoi vertex (and hence the tessellation) is degenerate if four edges are incident on
it, as in a simple cubic lattice of hard-disks, with square cells. A 3D Voronoi vertex is degenerate
if eight edges are incident on it, as in a face-centered cubic lattice of hard-spheres, with rhombic
dodecahedron cells.
Chapter 3 57
4 6 8 10 12 14
10
6
10
4
10
2
10
0
PSfrag replacements
P
n
Figure 3.3: P
n
for 2D Poisson tessellation. Theoretical result () from Calka (2003). Sim-
ulation data () averaged for 3500 frames of 900 random points.
gives,
Z = 1 +

v
=
v
v
=
1
1
. (3.10)
This is the Tonks equation of state for hard rods, Tonks (1936).
For hard rod system f
n+1
(r) can be got from f
n
(r) exactly as
f
n+1
(r) =
_
r
0
f
1
(r r
) f
n
(r
) dr
. (3.11)
Such a simple convolution is not available for hard-disk and hard-sphere systems.
3.4 Hard-disk results
For 2D Poisson tessellation f
1
(r) was derived by Collins (1968),
f
1
(r) =
r
3
_
1/2
r exp
_
4
r
2
_
+ erfc
_
2

1
2
r
__
. (3.12)
It was rederived by Stillinger et al. (2000) by a different method. Explicit expres-
sions for 2D Poisson tessellation P
n
are available in Calka (2003). Drouffe and
0 0.2 0.4 0.6 0.8 1
12
12.5
13
13.5
14
0.65 0.7 0.75 0.8 0.85
12
12.1
12.2
12.3
12.4
PSfrag replacements
C
2
Figure 3.4: C
2
for hard-disk thermodynamic () and swelled random structures at success
rates 10% (), 30% (), 50% (), 70% (+) and 90% (). NVE data averaged for 10000
con gurations of 256 hard-disks. Swelled random data averaged for 1000 con gurations of
256 hard-disks.
Itzykson (1984) and Tanemura (2003) report the simulation results for P
n
. We
compare the data of Calka (2003) with our simulation results in Fig. 3.3.
As mentioned in Section 3.2, for any nondegenerate 2D tessellation with PBC,
C
1
= 6 exactly. Hence, C
1
is not a microstructural indicator for hard-disk struc-
tures. However C
n
, for n > 1, are functions of and sensitive indicators of mi-
crostructure. C
2
and C
3
for hard-disk structures are given in Figs. 3.4 and 3.5, from
which we observe the following:
For 2D Poisson tessellation C
0
2
13.698 and C
0
3
22.94.
Well below the freezing density, the swelled random structures are identical
to the thermodynamic structures for any success rate. Above the freezing
density, the thermodynamic structure C
n
(n > 1) decreases sharply to the
regular hexagonal lattice values (C
n
)
reg
= 6n, while the swelled random
structure C
n
decrease slowly and nearly saturate at
DRP
.
Chapter 3 59
0 0.2 0.4 0.6 0.8 1
18
19
20
21
22
23
PSfrag replacements
C
3
Figure 3.5: C
3
rates 10% (), 30% (), 50% (), 70% (+) and 90% (). Averaging as in Fig. 3.4
If the swelled random congurations are generated with a low success rate ,
the large random trial displacements tend to equilibrate the local structures.
However, if success rate is high, the random trial displacements are small and
the swelling process locks the the particles into random structures. Fig. 3.4
inset shows that as the success rate is lowered, the C
2
of the resultant struc-
ture gets closer to its thermodynamic value. The difference between the C
n
(n ,= 1 in 2D) of a given structure and that of the thermodynamic structure
at the same density is a measure of its randomness. Fig. 3.4 inset shows that,
for >
F
, structures with different degrees of randomness can be gener-
ated by tuning the success rates. The proximity of the 70% and 90% success
rate structures in Fig. 3.4 inset, shows that the limiting case of near 100%
success rate should give the maximally random structures. These structures
should not sense the freezing transition, and hence their C
n
should not have
an inection point around the freezing density. From Fig. 3.4 inset, note that
while the 10% success rate structures have an inection about
F
, structures
with success rates 50% and above show no visible inection. The branch of
maximally random structures is presumed to terminate at the dense random
packing. However, by negotiating disorder with order, one can generate struc-
tures denser than the dense random packing, Torquato et al. (2000). From
Fig. 3.4 inset, note that at 10% success rate, packing fractions as high as 0.85
are attainable, even though
DRP
= 0.82 0.02. It is interesting to note that
the Voronoi neighbor statistics are sensitive even to the degree of randomness
of the random structures.
Next we study the number distribution of the Voronoi polygon edges, P
n
. For
2D congurations with PBC, even though C
1
=
n P
n
= 6 exactly, P
n
is a
function of density and a sensitive microstructural indicator. In Fig. 3.6 we compare
the hexagon incidence in the swelled random structures for different success rates
with that in the thermodynamic structures. However, for the incidence of other
polygons, to avoid a profusion of gures, we compare only the thermodynamic
structures (Fig. 3.7) and the 50% success rate swelled random structures (Fig. 3.8).
From these gures we observe the following:
Fig. 3.6 shows that, while the hexagon population rises sharply across the
freezing transition for thermodynamic structures, it rises quite slowly for the
random structures. As the success rate increases, the hexagon population at
DRP
decreases. This shows that increasing the success rate increases the
randomness of the structures.
From Fig. 3.7, for >
F
we see that the polygons dominant after hexagons,
are pentagons and heptagons. Also their populations are nearly identical. This
population-equality, also observed in the random structures Fig. 3.8, may be
explained as follows: In 2D structures with PBC, if the populations of the
polygons other than pentagons, hexagons and heptagons are negligible (as
in dense hard-disk structures), then the populations of pentagons and hep-
tagons will be nearly identical, so that the mean number of sides is exactly
six. Deng et al. (1989) (and references therein) show that defects in dense
two-dimensional systems could be represented by pairs of pentagon and hep-
tagons, and consider them to be a structural dipole.
Chapter 3 61
0 0.2 0.4 0.6 0.8
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0.9
1
PSfrag replacements
P
6
Figure 3.6: P
6
rates 10% (), 30% (), 50% (), 70% (+) and 90% (). Averaging as in Fig. 3.4
From Fig. 3.8, it is seen that the pentagon and heptagon populations are quite
signicant in the dense random hard-disk structures.
Polygons with faces 3, 10, 11 and 12 have sharply decreasing incidence as
increases (even for <
F
) in both thermodynamic and swelled random
structures, gure not shown.
3.5 Hard-sphere results
0
1
=
48
35
2
+2 15.5354, an exact result by Meijering
(1953). The Wigner-Seitz or Voronoi cell for the perfect fcc lattice is the rhombic
dodecahedron. For a perfect fcc lattice Fuller (1975) gives C
n
= 10 n
2
+ 2, that
is C
1
= 12 and C
2
= 42 and so on. Figs. 3.9 and 3.10 respectively show C
1
and
C
2
for thermodynamic and swelled random hard-sphere structures, from which we
observe the following:
0 0.2 0.4 0.6 0.8
10
6
10
4
10
2
10
0
PSfrag replacements
P
n
Figure 3.7: P
n
for hard-disk thermodynamic structures, for n =4 (), 5 (), 7 (), 8 (),
and 9 (+). Averaging as in Fig. 3.4.
0 0.2 0.4 0.6 0.8
10
6
10
4
10
2
10
0
PSfrag replacements
P
n
Figure 3.8: P
n
for hard-disk 50% success rate swelled random structures, for n =4 (), 5
(), 7 (), 8 (), and 9 (+). Averaging as in Fig. 3.4.
Chapter 3 63
0 0.2 0.4 0.6
14
14.2
14.4
14.6
14.8
15
15.2
15.4
15.6
PSfrag replacements
C
1
Figure 3.9: C
1
for hard-sphere thermodynamic (), 50% success rate swelled random ()
structures and () are experimental results by Finney (1970). The NVE and swelled random
data sets are averaged for 1000 con gurations of 256 hard-spheres.
0 0.2 0.4 0.6
50
55
60
65
70
PSfrag replacements
C
2
Figure 3.10: C
2
for hard-sphere thermodynamic () and 50% success rate swelled random
() structures. Averaging as in Fig. 3.9.
0
2
69.8.
The sudden decrease of C
n
across the freezing transition is similar to that
observed in the hard-disk system.
Finney (1970) has reported C
1
for two different sets of experimental dense
random packing congurations as 14.251 0.015 and 14.28 0.05. Fig. 3.9
shows that the C
1
for random hard-sphere structures match reasonably with
his experimental results.
In Fig. 3.9 shows that the hard-sphere solid structure C
1
approaches 14 instead
of 12. This phenomena is due to the topological instability of fcc lattice to
slight perturbations causing the coexistence of polyhedra with faces 12 to 18,
with the mean at 14. This was proved by Troadec et al. (1998), and for the
sake of completion, we outline the proof in Appendix 3.A. C
1
increases from
12 to 14 by the promotion of a few second neighbors into rst neighbors, by
the formation of additional tiny quadrilateral faces on the erstwhile rhombic
dodecahedron, see Fig. 3.19. This promotion manifests as a secondary feature
in g
1
(r) (Fig. 3.11) which grows as an inection near
F
, and becomes a
separate peak as increases.
Table 3.1 is the system size/shape dependence and thermodynamic consistency
checks on the hard-sphere C
1
data. It shows that C
1
tending to 14 near regular close
packing is consistent with the thermodynamic data. C
1
data shows negligible size
dependence since it depends only on the enumeration of the rst neighbors. For
simulations in a cubical box with PBC, the number of spheres must be more than
(6/) (C
1
+ + C
(n1)
) to have meaningful averages for the higher order C
n
.
Next we study the number distribution of the Voronoi polyhedra faces, P
n
. We
compare the data for the thermodynamic congurations in Figs. 3.12 and 3.13 with
those for the 50%success rate swelled randomcongurations in Figs. 3.14 and 3.15.
A rise in the dodecahedron population marks the freezing transition. However
the populations of 13 to 18 faceted polyhedra remain signicant even near
c
. Infact, near
c
, the population of 13 to 16 faceted polyhedra are more
Chapter 3 65
Table 3.1: System size/shape dependence and thermodynamic consistency checks for hard-
sphere C
1
.
Z
HS
a
Run C
1
f
1
() Z
vor
b
0.65 24.19 I
c
14.0390 9.96 24.31
II
d
14.0388 9.99 24.38
III
e
14.0386 9.83 23.99
IV
f
14.0389 9.92 24.20
V
g
14.0382 9.93 24.24
0.68 36.34 I 14.0252 15.00 36.07
II 14.0260 15.13 36.37
III 14.0251 15.30 36.77
IV 14.0255 15.09 36.27
V 14.0249 15.20 36.53
0.70 54.47 I 14.0172 22.97 54.65
II 14.0168 23.08 54.93
III 14.0165 23.11 55.00
IV 14.0164 23.16 55.10
V 14.0175 22.97 54.67
0.72 108.05 I 14.0083 46.23 108.94
II 14.0080 46.15 108.74
III 14.0083 46.67 109.95
IV 14.0083 46.29 109.08
V 14.0085 46.63 109.86
a
Young and Alder (1979) give the hard-sphere solid equation of state as Z = 3/+2.566+0.55
1.19
2
+ 5.95
3
, where = (v v
c
)/v
c
= (1 y)/y, is the dimensionless excess free-volume.
b
Using Eq. 3.8.
c
Averaged for 1000 con gurations of 256 hard-spheres in a cubic box, with PBC.
d
Averaged for 512 con gurations of 500 hard-spheres in a cubic box.
e
Averaged for 500 con gurations of 512 hard-spheres in a cuboidal box (l
x
: l
y
: l
z
= 2 : 1 : 1).
f
g
1 1.5 2 2.5
0
1
2
3
4
5
PSfrag replacements
r/
g
n
(
r
)
g
1
g
2
g(r)
Figure 3.11: g
n
(r) for thermodynamic hard-sphere structures at = 0.57. The promotion
of a few second neighbors into rst neighbors manifests as a secondary peak in g
1
(r).
than the dodecahedron population. Also P
n
near regular close packing is
nearly identical to that near the dense random packing, due to the topological
instability, Fig. 3.16. The P
n
data near dense random packing matches well
with that of Jullien et al. (1996).
Fig. 3.14 shows that in the random hard-sphere structures, the dodecahedron
population strangely decreases with increasing . This behavior is unlike that
in the random hard-disk structures (Fig. 3.6), where the hexagon population
steadily increases with . The decrease in dodecahedron population cannot
be interpreted as a decrease in fcc crystallites, because a slightly perturbed fcc
crystallite may get accounted in the 13 to 18 faceted polyhedra population.
Comparing the P
n
data in Figs. 3.13 and 3.15, we see that the population
of polyhedra with faces 11, 16 to 19 decrease sharply across the freezing
transition in the thermodynamic structures, but they persist signicantly in
the random structures.
Chapter 3 67
0 0.2 0.4 0.6
10
2
10
1
10
0
PSfrag replacements
P
n
Figure 3.12: P
n
for hard-sphere thermodynamic structures, for n =12 (), 13 (), 14 (),
and 15 (). Averaging as in Fig. 3.9.
0 0.2 0.4 0.6
10
5
10
4
10
3
10
2
10
1
10
0
PSfrag replacements
P
n
Figure 3.13: P
n
for hard-sphere thermodynamic structures, for n =11 (), 16 (), 17 (),
18 (), and 19 (+). Averaging as in Fig. 3.9.
0 0.2 0.4 0.6
10
2
10
1
10
0
PSfrag replacements
P
n
Figure 3.14: P
n
for hard-sphere 50% success rate swelled random structures, for n =12
(), 13 (), 14 (),and 15 (). Averaging as in Fig. 3.9.
0 0.2 0.4 0.6
10
5
10
4
10
3
10
2
10
1
10
0
PSfrag replacements
P
n
Figure 3.15: P
n
for hard-sphere 50% success rate swelled random structures, for n =11
(), 16 (), 17 (), 18 (), and 19 (+). Averaging as in Fig. 3.9.
Chapter 3 69
10 12 14 16 18 20 22
0
0.05
0.1
0.15
0.2
0.25
0.3
0.35
0.4
PSfrag replacements
n
P
n
Figure 3.16: Comparison of P
n
for near regular close packing (thermodynamic structure) at
= 0.74 () and near dense random packing (swelled random structure) at = 0.632 ().
A similar topological instability occurs in the hard-disk system also. A simple
cubic lattice of hard-disks has square Voronoi polygons, hence C
1
= 4. In 2D
tessellations, vertices with four edges incident on them are topologically unstable
and any slight perturbation of the lattice renders C
1
= 6. However, since the regular
close-packed structure in two dimensions is a hexagonal structure, in which the
number of Voronoi edges is stable under a slight perturbation of the particle centers,
this effect is not observed in the two-dimensional solid structures.
7
0
V
o
r
o
n
o
i
n
e
i
g
h
b
o
r
s
t
a
t
i
s
t
i
c
s
o
f
h
a
r
d
-
d
i
s
k
s
a
n
d
h
a
r
d
-
s
p
h
e
r
e
s
Table 3.2: Parameters
0
and T
0
in
0
[1 +(T
/T
0
)
1/2
]
1/6
for the different EHSD models, as presented in Ben-Amotz and Herschbach
(1990).
Model
0
T
0
Barker Henderson (BH) 1.1154 1.759
Weeks Chandler Anderson (WCA) 1.1137
_
0.72157 + 0.04561
0.07468
2
+ 0.12344
3
2
Lado (LWCA) 1.1152
_
0.73454 + 0.10250
0.12960
2
+ 0.15976
3
2
Ben-Amotz Herschbach (BAH) 1.1532 0.527
Chapter 3 71
3.6 Estimation of the effective hard-sphere diameter for Lennard-
Jones uid fromC
1
The notion of EHSD has a long history, beginning with Boltzmann. He suggested
that the distance of closest approach of the soft potential molecules could be con-
sidered the EHSD, Boltzmann (1964). Experiments and simulations have shown
that the structure, characterized by the radial distribution function (Kirkwood and
Boggs (1942)) or its Fourier transform, the structure factor (Verlet (1968)), of a
dense soft potential uid can be matched with that of the hard-sphere uid having
a particular diameter. This denes the EHSD of the soft potential uid at the given
density and temperature. The EHSD method has proven successful in the predic-
tion of thermodynamic properties, self-diffusion coefcient (Silva et al. (1998)) and
shear viscosity (Heyes (1988)) of the Lennard-Jones (LJ) uids.
The LJ potential
LJ
(r) = 4
LJ
[(
LJ
/r)
12
(
LJ
/r)
6
], has a energy scale
LJ
and a length scale
LJ
called the molecular diameter. The dimensionless tempera-
ture is T
= k
B
T/
LJ
and the dimensionless density is
=
3
LJ
. The effective
hard-sphere diameter is rendered dimensionless as
= /
LJ
. The theoretical
approaches due to Barker and Henderson (1976) (BH), Weeks-Chandler-Andersen
(Chandler et al. (1983)) (WCA), and its modication by Lado (1984) (denoted as
LWCA), integrate the repulsive part of the LJ potential with different criteria yield-
ing different expressions for the EHSDs. The explicit expressions for these models,
as presented in Ben-Amotz and Herschbach (1990), along with their empirical result
(denoted here as BAH) got by tting the equation of state data to Carnahan-Starling-
van der Waals equation are given in Table 3.2.
We estimate the EHSD as follows. By Voronoi tessellating the LJ congurations
we get C
1
. The effective hard-sphere packing fraction is got by interpolating the
thermodynamic hard-sphere C
1
vs data, Fig. 3.9. Then, the dimensionless EHSD
is computed as
= (6 //
)
1/3
. This method of estimating the EHSD of a
soft potential uid requires that the averaged local neighborhood, as characterized
by C
1
, be identical with that of the hard-sphere uid having the diameter , the
EHSD. This method is based on the statistical geometry of particle distributions
and is similar to the Boltzmanns method which is based on the trajectories. Such
Table 3.3: Comparison of EHSD values predicted from C
1
with those from the models in Table
3.2.
LJ state
values from the models in Table 3.2 Voronoi analysis

T
BH WCA LWCA BAH C

1
a
0.7408 0.8350 1.0262 1.0224 1.0200 1.0123 14.47 1.0318

0.8230 0.8010 1.0226 1.0193 1.0169 1.0074 14.55 1.0295
1.0649 0.7000 1.0134 1.0113 1.0090 0.9952 14.77 1.0246
1.0662 0.8210 1.0133 1.0096 1.0068 0.9951 14.58 1.0143
1.0845 0.7690 1.0127 1.0097 1.0072 0.9943 14.67 1.0188
2.5655 0.4000 0.9775 0.9782 0.9760 0.9497 15.24 0.9884
2.7371 0.3000 0.9746 0.9758 0.9739 0.9461 15.33 0.9830
2.7584 0.7195 0.9742 0.9718 0.9686 0.9457 14.91 0.9735
3.2617 0.9200 0.9666 0.9600 0.9558 0.9364 14.70 0.9540
3.8833 0.9900 0.9583 0.9496 0.9449 0.9267 14.66 0.9398
a
Averaged for 1000 con gurations of 256 LJ molecules, with PBC.
statistical geometric approaches are elegant because they do not employ any abstract
integral criteria for the repulsive part of the potential (as in BH, WCA or LWCA
models) or appeal to any property data tting (as in BAH model). Table 3.3 shows
10 different state points for the LJ uid, the rst 5 state points are in liquid state,
while the rest are in gaseous state. For these state points, the values of EHSD
predicted from C
1
show less than 2% deviation from the BH, WCA and LWCA
models and less than 5% deviation from the BAH model. It may be noted that the
deviations among these models is also of the same order (see for example, Fig.7 of
Ben-Amotz and Herschbach (1990)). The excellent match of the EHSD computed
from the statistical geometric approach with those based on the integral criteria
for repulsive part of the potential (as in BH, WCA and LWCA models) shows the
validity of the EHSD concept and also acts as a validation of the computational
procedure used here.
Chapter 3 73
3.7 Conclusions
We have analyzed the C
n
and P
n
for the thermodynamic and swelled random struc-
tures, for the entire density range of hard-disks and hard-spheres. Well below the
freezing density, the random structures produced at any success rate are identical
to the thermodynamic structures. Above the freezing density, higher success rates
produce more random structures, and the limit of near 100% success rate gives the
maximally random structures. These maximally random structures should not sense
the freezing transition, and hence their C
n
should not have an inection point around
the freezing density. The branch of maximally random structures is presumed to ter-
minate at the dense random packing. However, by negotiating disorder with order,
we are able to generate structures denser than the dense random packing. C
1
for
random hard-sphere structures produced by our algorithm matches reasonably with
the dense random packing experimental results of Finney (1970).
For hard rod system g
n
(r) is exactly known, Fisher (1964). For 2D Poisson tes-
sellation f
1
(r) (Collins (1968)) and P
n
(Calka (2003)) are exactly known. For any
nondegenerate 2D tessellation with periodic boundary conditions, C
1
= 6 exactly,
Meijering (1953); Collins (1968). For the 2D Poisson tessellation, we report C
0
2

13.698 and C
0
3
22.94. For the 3D Poisson tessellation, C
0
1
=
48
35
2
+2 15.5354
exactly (Meijering (1953)), and we report C
0
2
69.8.
On freezing, the hard-disk C
n
(for n > 1) decrease sharply to the regular hexag-
onal lattice values (C
n
)
reg
= 6n. A sharp rise in the hexagon population, and
a sharp drop in the population of the other polygons, mark the onset of hard-disk
freezing transition. The hard-disk random structures have a slow rise in the hexagon
population with increasing density, and the pentagon and heptagon populations re-
main signicant. In dense hard-disk structures, both thermodynamic and random,
the pentagon and heptagon populations are nearly identical.
For the perfect fcc lattice C
1
= 12, and C
2
= 42. However, due to topolog-
ical instability (Troadec et al. (1998)), a slightly perturbed fcc lattice has Voronoi
polyhedra with faces 12 to 18, with the mean at 14. This increase is achieved by
forming tiny quadrilateral faces with a few second neighbors, thereby promoting
them into rst neighbors. This promotion manifests as a secondary peak in g
1
(r).
Thus, on freezing transition the hard-sphere C
1
is close to 14 rather than 12. We
have demonstrated that this result is consistent with thermodynamic data. On freez-
ing transition, even though there is a rise in the rhombic dodecahedron population,
the population of polyhedra with 13 to 18 faces remain signicant. In hard-sphere
random structures, the dodecahedron population decreases with increasing density.
These results show the signicant differences between the hard-sphere and hard-
disk microstructures.
We show that the Voronoi neighbor statistic C
1
is useful in estimating the ef-
fective hard-sphere diameter of soft potential uids. By matching the C
1
of the
Lennard-Jones uid congurations with that of the thermodynamic hard-sphere
uid at some packing fraction, we are able to estimate the effective hard-sphere di-
ameter of the Lennard-Jones uid within 2% deviation from the theoretical results
for Barker-Henderson, Weeks-Chandler-Andersen and its modication by Lado.
This statistical-geometric approach is elegant because it does not employ any ab-
stract integral criteria for the repulsive part of the potential (as in the above men-
tioned theories) or appeal to any property data tting (as in the empirical correla-
tions).
Appendix
3.A Topological instability of fcc lattice
The Rhombic dodecahedron (Fig. 3.17) is the Wigner-Seitz cell or the Voronoi poly-
hedron for the perfect fcc lattice. It has 12 identical rhombic faces and 14 vertices.
Its vertices are classied into two types, based on their connectivity. Vertex type-A
has four edges incident on it, with three edges from the given cell and another edge
from the neighboring cells. Vertex type-B has eight edges incident on it, with four
edges from the given cell and four from the neighboring cells. A rhombic dodeca-
hedron has eight type-A vertices and six type-B vertices. A type-B vertex is shared
by six spheres, the centers of which form an octahedron, Fig. 3.18. The type-B ver-
tices are topologically unstable, and on perturbation form additional edges leading
to pentagonal or hexagonal faces or form an additional tiny quadrilateral face with a
second neighbor sphere, thereby promoting it into a rst neighbor, Fig. 3.19. When
Chapter 3 75
Figure 3.17: Rhombic dodecahedron, the Wigner-Seitz or Voronoi cell for the perfect fcc
lattice.
an additional face is formed, it is perpendicular to the diagonal of the octahedron
formed by the central sphere and the sphere getting promoted as a rst neighbor.
Among all possible innitesimal perturbations of the lattice shown in Fig. 3.18,
there will be equal number of perturbations leading to the formation of additional
faces between the pairs of spheres (1,6) or (2,4) or (3,5). Hence the probability that
an additional face is formed between one of the pairs is
1
3
. Thus, the average number
of faces = 12 + number of type-B vertices probability that a type-B vertex forms
an additional face = 12 + 6
1
3
= 14. This was proved by Troadec et al. (1998),
further details are therein.
1
2
3
4
5
6
Figure 3.18: The octahedron formed by the spheres sharing a type-B vertex. For sphere 1,
the spheres 2 to 5 are rst neighbors, while sphere 6 is the second neighbor, if the lattice is
nonperturbed. Small perturbations can promote sphere 6 into a rst neighbor for sphere 1.
Figure 3.19: A Voronoi cell in a perturbed fcc lattice, having pentagonal or hexagonal faces
due to the formation of additional edges or having an additional face formed with an earlier
second neighbor.
Chapter 4
Voronoi neighbor statistics of sheared
inelastic hard-disks and hard-spheres
In this chapter we extend the neighbor analysis to the homogeneously sheared in-
elastic hard-particle structures, the simplest model for rapid granular matter. The
pair distribution function is partitioned into the nth neighbor coordination num-
ber (C
n
), and the nth neighbor position distribution [f
n
(r)]. The distribution of
the number of Voronoi faces (P
n
) is also considered since C
1
is its mean. We re-
port the C
n
and P
n
for the homogeneously sheared inelastic hard-disk and hard-
sphere structures. These statistics are sensitive to shear ordering transition, the non-
equilibrium analogue of the freezing transition. In the near-elastic limit, the sheared
uid statistics approach that of the thermodynamic uid. On shear ordering, due to
the onset of order, the sheared structure C
n
drop to the thermodynamic solid phase
value. The suppression of nucleation by the homogeneous shear is evident in these
statistics. As inelasticity increases, the shear ordering packing fraction increases.
In shear ordered inelastic hard-sphere structures there is a high incidence of 14-
faceted polyhedra and a consequent depletion of polyhedra with faces 12, 13, 15
to 18, due to the formation of body-centered tetragonal (bct) structures. These bct
structures leave a signature like the body-centered cubic structure in the C
n
and P
n
data. On shear ordering, close-packed layers slide past each other. However, with
a velocity-dependent coefcient of restitution, at a critical shear rate these layers
get disordered or amorphized. We nd that the critical shear rate for amorphiza-
77
78 Voronoi neighbor statistics of sheared inelastic hard-disks and hard-spheres
tion is inversely proportional to the particle diameter, as compared to the inverse
square scaling observed in dense colloidal suspensions. These results are reported
in Kumar and Kumaran (2006b).
4.1 Introduction
In Chapter 3, we presented the Voronoi neighbor analysis for thermodynamic and
quenched structures. Here, we extend the analysis to homogeneously sheared hard-
disk and hard-sphere structures (Section 1.3), and compare them with the thermo-
dynamic (NVE-MC) hard-structures. Inelastic hard-particles are the simplest model
for rapid granular matter, Section 1.4. In most part of the chapter, we use the con-
stant coefcient of restitution (COR) model for inelastic collisions. Due to a lack
of an intrinsic energy scale, these structures are shear rate independent. Hence, to
study the effect of shear rate on structures we use the velocity-dependent COR given
in Eq. 1.5.
In Section 4.2 we introduce the Voronoi neighbor statistics. Let a central spheres
geometric neighbors be called rst neighbors, i.e. the rst layer of neighbors. The
rst neighbors neighbors (which are themselves not rst neighbors) are the second
neighbors, and so on. Thus, all the spheres surrounding a central sphere are par-
titioned layer-wise, and characterized by nth neighbor coordination number (C
n
)
and nth neighbor position distribution function [f
n
(r)]. In Section 4.2 we show
that the total information contained in the pair distribution function, g(r), can be
partitioned into these sets of neighbor statistics. The distribution of the number
of Voronoi bounding surfaces (P
n
) is also of interest, since C
1
is its mean. These
neighbor statistics are sensitive to the shear ordering transition. In this thesis, the
usage shear ordering is exclusively used for that non-equilibrium structural tran-
sition which approaches the thermodynamic ordering or freezing transition, in the
near-elastic limit. We avoid the usage shear-induced ordering, since homogeneous
shear in fact suppresses the crystallite nucleation, as discussed below.
In Section 4.3 and 4.4 we present the C
n
and P
n
for sheared inelastic hard-
disk and hard-sphere structures respectively. In the near-elastic limit the sheared
uid statistics approach that of the thermodynamic uid values. When the sheared
Chapter 4 79
neighbor statistics approach the thermodynamic solid phase values, shear order-
ing transition takes place. On shear ordering, close packed strings of hard-disks
slide over each other in the velocity direction, as observed in the sheared mono-
layer experiments of Stancik et al. (2003). The dislocations in a crystalline solid
do not move with the same degree of ease along different crystallographic planes.
Generally there are preferred crystallographic planes called the slip planes, and the
preferred directions lying in those planes are called the slip directions. For the face-
centered cubic (fcc) lattice, the 111 family of planes are the slip planes and the
110) directions are the slip directions. A slip plane and a slip direction constitute
a slip system. Materials plastically deform by slipping along a slip system, Callis-
ter Jr. (2000). In dense sheared hard-sphere simulations, having an fcc lattice as the
initial conguration, an orientation is said to be preferred if the velocity-vorticity
plane and the velocity vector form a slip system, any other orientation is said to be
unpreferred. In the preferred orientation shear of inelastic hard-spheres, on shear
ordering, close packed two-dimensional layers of spheres slide over each other in a
zig-zag path, as observed in plastic deformation of fcc crystals (Cottrell (1953)) and
colloidal suspensions (Ackerson (1990)). In colloids and dense suspensions, such a
layer-wise ow causes shear-thinning, since sliding close-packed layers offer lesser
resistance to shear than a disordered structure, Woodcock (1985). The Voronoi
statistics clearly show the suppression of crystal nucleation by homogeneous shear,
observed in the Brownian dynamics simulations of colloidal suspensions by Butler
and Harrowell (1995) and Blaak et al. (2004). However, in some experiments it is
observed that the in-plane ordering, caused by shear, increases the nucleation, see
the references in Blaak et al. (2004).
For a perfect fcc lattice C
1
= 12, since the unit cell is a rhombic dodecahe-
dron. However, an isotropically perturbed fcc lattice has polyhedra with faces 12
to 18, with a mean at 14, due to a topological instability discovered by Troadec
et al. (1998), we have outlined the original proof in the appendix of Chapter 3.
In shear ordered inelastic hard-sphere structures there is a high incidence of 14-
faceted polyhedra and a consequent depletion of polyhedra with faces 12, 13, 15
to 18. This shows the presence of a nearly body-centered cubic structure, since the
Voronoi polyhedron of the body-centered cubic (bcc) lattice is the truncated octa-
hedron, which has 14 faces. It is known that in the martensitic transformations, a
rapidly quenched fcc metallic structure gets sheared to the bct structure, Callister Jr.
(2000). The bct structures generated in a dense system of homogeneously sheared
inelastic hard-spheres leave a bcc-like signature in the C
n
and P
n
data.
In an unpreferred orientation shear of inelastic hard-spheres, at packing frac-
tions above shear ordering, the structures get disordered or amorphized. Such a
metastable amorphization is observed by Ikeda et al. (1999) in the unpreferred ori-
entation shear of nanocrystals. This is a metastable amorphization, since the sys-
tem will eventually recrystallize in the preferred orientation. When the imposed
displacement or velocity does not lie on a slip plane, simultaneous slip and shear
can break the lattice. We consider this to be the cause of amorphization in an un-
preferred orientation shear. These amorphized structures are nearly identical to the
dense random structures, except for the higher incidence of some high-faceted poly-
hedra at the expense of some low-faceted ones. This can be rectied by a structural
relaxation using the swelled random algorithm. Thus, the Lees-Edwards boundary
condition simulation for nearly elastic hard-spheres sheared in an unpreferred ori-
entation, followed by a structural relaxation using the swelled random algorithm,
offers a nearly deterministic algorithm to generate dense random packings.
A steady state amorphization is the cause of shear thickening observed in dense
suspensions, Hoffman (1972, 1998). The constant COR sheared structures are shear
rate independent, hence they cannot amorphize as shear rate is varied. We showthat,
with a velocity-dependent COR, steady state amorphization is achieved even in the
preferred orientation. Even when the system is globally in the preferred orientation,
the slip planes are locally mis-oriented near edge dislocations and grain boundaries.
At these locations, shear induces local amorphization, since velocity does not lie
on these local slip planes. When the local amorphization rate exceeds the local
recrystallization rate, a steady amorphized state results. We observe that the critical
shear rate for amorphization is inversely proportional to the sphere diameter, as
predicted by Savage and Jeffrey (1980) for granular suspensions, unlike the inverse
square scaling observed in the colloidal suspensions, Hoffman (1972, 1998).
Chapter 4 81
4.2 Voronoi partitioning of the pair distribution function
Let N
n
be the number of nth neighbors around a central disk, then the nth neigh-
bor coordination number is C
n
= N
n
), where ) denotes time averaging. The
two-dimensional (in the most general case) anisotropic pair distribution function is
computed as
g(r, ) =
(r, )
=
1
N
r,
)
V
r,
, (4.1)
where V
r,
is the area element r dr d about (r, ) and N
r,
is the number of disks
with their centers in that area element. The disks around the central disk can be
partitioned layer-wise as
N
r,
=
n=1
N
n
r,
, (4.2)
where N
n
r,
is the number of nth neighbors with their centers in the area element
dV
r,
. Using Eq. 4.2 in Eq. 4.1, g(r, ) can be partitioned as
g(r, ) =
1
n=1
N
n
r,
_
V
r,
=
n=1
g
n
(r, ), (4.3)
where g
n
(r, ) is the nth neighbor anisotropic pair distribution function. Such a
partitioning for the isotropic uid was rst reported by Rahman (1966).
The nth neighbor anisotropic position distribution function, f
n
(r, ), is dened
such that f
n
(r, ) r dr d is the fraction of the nth neighbors in the area segment
r dr d about (r, ). Then
f
n
(r, ) r dr d =
N
n
r,
_
_
_
r=0
_
2
=0
N
n
r,
r dr d
_ =
N
n
r,
_
C
n
. (4.4)
Here, we have used
_
_
r=0
_
2
=0
N
n
r,
r dr d
_
= N
n
) = C
n
. Using Eq. 4.3 and
Eq. 4.4, we get,
g
n
(r, ) =
C
n
f
n
(r, ). (4.5)
Here, we have used V
r,
= r dr d. Then
g(r, ) =
1
n=1
C
n
f
n
(r, ). (4.6)
The analogous result for three-dimensional anisotropic pair distribution function is
g(r, , ) =
1
n=1
C
n
f
n
(r, , ). (4.7)
Here, f
n
(r, , ) is dened such that f
n
(r, , ) r
2
sin dr d d is the fraction of
the nth neighbors in the volume element r
2
sin dr d d about (r, , ), where
0 < 2 and 0 . Equations 4.6 and 4.7 show that the two Voronoi
neighbor statistics, C
n
and f
n
(r) together contain the structural information in the
anisotropic pair distribution function g(r). We consider another Voronoi statistic,
the distribution of the number of bounding surfaces of the Voronoi cell, P
n
. It
is identical to the distribution of the number of the rst neighbors, hence C
1
=
nP
n
. In Section 4.3 and 4.4, we show that these neighbor statistics are sensitive
indicators of the microstructural changes occurring in sheared hard-disk and hard-
sphere structures.
4.3 Hard-disk results
For any non-degenerate two-dimensional tessellation with periodic boundary condi-
tions (PBC) or with a large number of particles, C
1
= 6 exactly, Meijering (1953);
Collins (1968). Hence we present the values of C
2
and C
3
for the entire density
range, at different values of , in Figs. 4.1 and 4.2 respectively. From these gures
we observe the following:
In the near-elastic limit, the sheared uid structures are nearly isotropic. Hence
the sheared uid structure C
n
(in Figs. 4.1 and 4.2) and the angular-averaged
pair distribution function (in Fig. 4.3)
g(r) =
1
2
_
=2
=0
g(r, ) d,
approach their respective thermodynamic uid values. Since the neighbor
statistics are by denition angular-averaged, the angular-averaged pair distri-
bution function is more appropriate for our analysis than the anisotropic pair
distribution function, even though the latter contains the complete structural
information.
Chapter 4 83
At packing fractions above the freezing packing fraction (
F
0.691, Alder
and Wainwright (1962)), the thermodynamic hard-disk C
n
reduces sharply to
its regular hexagonal lattice values (C
n
)
reg
= 6n. Above the melting packing
fraction (
M
0.716, Alder and Wainwright (1962)) the hard-disk system
exists as the ordered or solid phase.
When the sheared structure C
n
approaches that of the thermodynamic solid
phase C
n
, shear ordering transition takes place. For example, for = 0.99
the ordering transition occurs near = 0.80 as signaled in Figs. 4.1 and
4.2. This criterion is cross-checked with the g(x, y) plots in Fig. 4.4, for
= 0.75 and = 0.80. On shear ordering, strings of disks slide over each
other along the velocity direction, as shown by the tails of the bright spots in
Fig. 4.4 for = 0.80. Shear of monolayers of polystyrene spheres suspended
at the decane-water interface show that particle strings slide past each other
at high shear rates, Stancik et al. (2003). Thus the C
n
are sensitive to both
the thermodynamic and shear ordering transitions. The sheared hard-disk C
n
show negligible system size dependence, as observed from Table 4.1. Since
the C
n
are the angular-averaged constituents of g(r, ) (Eq. 4.6), the formers
system size dependence will be comparable, if not less due to the angular
averaging, than the latter.
The Brownian dynamics simulations of colloidal suspension by Butler and
Harrowell (1995) and Blaak et al. (2004) show that homogeneous shear sup-
presses the crystal nucleation. In the inset of Fig. 4.1, note that even though
the near-elastic ( = 0.999) sheared uid structure C
2
matches with the ther-
modynamic uid structure C
2
till
F
, for >
F
the sheared structure C
2
drops sluggishly, while the thermodynamic structure C
2
drops steeply. This
indicates the suppression of crystal nucleation by shear.
The inset of Fig. 4.1 shows that as inelasticity increases (i.e. decreases)
shear ordering occurs at a higher packing fraction. Assuming local equi-
librium, this phenomenon can be reasoned in analogy with thermodynamic
melting as follows: Clausius-Clapeyron equation shows that the melting tem-
perature of a molecular system decreases with decreasing pressure, provided
the molar volume of the solid phase is lower than that of the liquid phase
and latent heat of fusion is positive. A decrease in melting temperature in a
molecular system, translates to an increase in melting density for an athermal
system. At a given , as inelasticity increases granular pressure decreases.
Then, upto an isotropic approximation, uidity prolongs i.e. the shear order-
ing packing fraction increases.
In Figs. 4.1 and 4.2, the data for = 0.70 and 0.50 terminate without touch-
ing the thermodynamic solid phase due to dynamic inelastic collapse, Alam
and Hrenya (2001). Dynamic inelastic collapse is a pathological behavior
in event-driven simulations where an innite number of collisions occur in
a nite time within clusters, signaled by a sudden drop in collision times
and hence negligible evolution of the system in real time. This can be cir-
cumvented by inducing a homogeneous density eld using impact velocity
dependent coefcients of restitution, Poschel et al. (2003). Even though we
simulated such a model to extend the data of = 0.70 and 0.50, we do not
present it in Figs. 4.1 and 4.2 to maintain the simplicity of two parameter
analysis ( and ).
It is interesting to note that the trends in C
2
and C
3
identical, even though
they characterize the second and third neighbors respectively. This shows
that Voronoi partitioning is a physically relevant parametrization.
Next, we study the number distribution of the Voronoi polygon edges, P
n
. For
two-dimensional congurations with PBC, even though C
1
=
nP
n
= 6 exactly,
P
n
is a function of density and a sensitive microstructural indicator. We compare
the P
n
data of the thermodynamic and sheared structures in Figs. 4.5, 3.7 and 4.6,
from which we observe the following:
In two-dimensional systems hexagonal structure is the regular close packing.
In thermodynamic structures, across the freezing transition, in response to the
onset of order, there is a sharp rise in the hexagon incidence, Fig. 4.5. Con-
sidering the approach of sheared structure P
6
to that of the thermodynamic
structures as the sign of shear ordering leads to conclusions same as before:
In the near-elastic limit, the sheared uid P
n
approaches that of the thermo-
dynamic uid. The suppression of crystal nucleation by shear manifests as
Chapter 4 85
0 0.2 0.4 0.6 0.8
12
12.5
13
13.5
14
0.65 0.7 0.75 0.8 0.85
12
12.1
12.2
12.3
12.4
12.5
PSfrag replacements
C
2
Figure 4.1: C
2
for sheared inelastic hard-disk structures at = 0.50 (), 0.70 (), 0.90 (),
0.95 (+), 0.99 () and 0.999 (), compared with thermodynamic structures (). All data
averaged for 10000 con gurations of 256 hard-disks, with PBC.
0 0.2 0.4 0.6 0.8
18
19
20
21
22
23
PSfrag replacements
C
3
Figure 4.2: C
3
0.95 (+), 0.99 () and 0.999 (), compared with thermodynamic structures (). Averag-
ing as in Fig. 4.1.
1 1.5 2 2.5 3
0.5
1
1.5
2
2.5
3
3.5
4
4.5
1.4 1.5 1.6 1.7 1.8 1.9 2 2.1 2.2
0.7
0.8
0.9
1
1.1
1.2
1.3
1.4
PSfrag replacements
r/
g
(
r
)
Figure 4.3: g(r) at = 0.50 for sheared inelastic hard-disk structures at = 0.50 (dotted
line), 0.70 (thin dashed line), 0.90 (thin line), 0.95 (thick dashed dotted line) and 0.99 (thick
dashed line), compared with thermodynamic structure (thick continuous line). Averaged
over 10000 con gurations of 256 hard-disks with PBC.
4 2 0 2 4
4
3
2
1
0
1
2
3
4
PSfrag replacements
x/
y
/
4 2 0 2 4
4
3
2
1
0
1
2
3
4
PSfrag replacements
x/
y/
x/
y
/
Figure 4.4: Pair distribution function for sheared inelastic hard-disks at = 0.75 (left) and
= 0.80 (right), at = 0.99. Averaged over 10000 con gurations of 256 hard-disks, with
PBC.
Chapter 4 87
Table 4.1: System size dependence check for sheared inelastic hard-disk structure C
2
and C
3
,
for a constant coef cient of restitution = 0.99.
Run C
2
C
3
0.70 I
a
12.2883 18.7492
II
b
12.2880 18.7486
III
c
12.2875 18.7483
IV
d
12.2878 18.7485
0.75 I 12.1467 18.3918
II 12.1457 18.3895
III 12.1534 18.4063
IV 12.1393 18.3666
0.80 I 12.0159 18.0431
II 12.0279 18.0767
III 12.0679 18.1864
IV 12.0614 18.1616
0.83 I 12.0067 18.0186
II 12.0077 18.0214
III 12.0204 18.0581
IV 12.0086 18.0223
0.84 I 12.0049 18.0135
II 12.0058 18.0161
III 12.0051 18.0137
IV 12.0049 18.0126
a
Averaged for 10000 con gurations of 256 hard-disks, l
x
: l
y
= 1 :
3/2.
b
x
: l
y
= 1 :
3/2.
c
x
: l
y
= 1 :
3/2.
d
x
: l
y
= 1 :
3/2.
the sluggish rise of the near-elastic (for = 0.999) sheared structure P
6
for
>
F
, as compared to the steep rise of the thermodynamic P
6
. As inelastic-
ity increases the shear ordering packing fraction increases.
From Fig. 3.7 note that in thermodynamic structures, the incidence of poly-
gons with edges 4, 8 and 9 decreases sharply after the freezing transition and
the defects in the hard-disk solid structures are represented by nearly identi-
cal populations of pentagons and heptagons. Deng et al. (1989) (and refer-
ences therein) show that defects in dense two-dimensional systems could be
represented by pairs of pentagon and heptagons, and consider them to be a
structural dipole. Comparing Figs. 3.7 and 4.6, note that the sharp drop in
the incidence of polygons with edges 4, 8 and 9 occurs near the shear order-
ing packing fraction, which is greater than the freezing packing fraction due
to inelasticity. The incidence of pentagons and heptagons in dense sheared
structures are much greater than those in the dense thermodynamic structures.
The incidence of polygons with edges 3, 10, 11 and 12 decreases sharply with
increasing density even for <
F
(gure not shown).
4.4 Hard-sphere results
Solids with fcc structure undergo plastic deformation by slipping along the < 110 >
directions in the 111 planes, Callister Jr. (2000). Experimental studies show that
in a plane Poiseuille ow of colloidal crystals, the crystals are aligned with their
(111) plane parallel to the cell wall and the [110] direction along the velocity, Kanai
et al. (2003). For a recent review on ow-induced structure in colloidal suspensions
see Vermant and Solomon (2005). An orientation in which a 111 plane parallel
to the velocity-vorticity (xz) plane is the preferred orientation for the shear of fcc
lattice, and the other orientations are unpreferred. The usual way of generating the
initial fcc conguration for molecular dynamics simulations (Allen and Tildesley
(1992)) is an unpreferred orientation, since the (001) plane is parallel to the xz
plane. In a sheared uid state, the orientation of the initial conguration does not
matter (shown below). It is known the colloidal solids on shearing in an unpreferred
Chapter 4 89
0 0.2 0.4 0.6 0.8
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0.9
1
0.65 0.7 0.75 0.8 0.85
0.6
0.7
0.8
0.9
1
PSfrag replacements
P
6
Figure 4.5: P
6
0.95 (+), 0.99 () and 0.999 (), compared with thermodynamic structures (). Averag-
ing as in Fig. 4.1.
0 0.2 0.4 0.6 0.8
10
6
10
4
10
2
10
0
PSfrag replacements
P
n
Figure 4.6: P
n
for sheared inelastic hard-disk structures at = 0.90, for n =4 (), 5 (), 7
(), 8 (), and 9 (+). Averaging as in Fig. 4.1.
orientation, melt and recrystallize along the preferred orientation. However, this
lattice reorientation requires long simulation times. Hence we generate initial con-
gurations in the preferred orientation by stacking layers of two dimensional close-
packed spheres in ABCABC . . . order. Suppose we have p stacks (along the y di-
rection) of q q spheres forming two-dimensional close-packed spheres (parallel to
xz plane), then the box relative dimensions are l
x
: l
y
: l
z
= q : p
_
2/3 : q
3/2.
Figs. 4.7 and 4.8 respectively compare the C
1
and C
2
of inelastic hard-spheres
sheared in the preferred orientation with those of the thermodynamic structures.
The C
n
are sensitive to thermodynamic freezing and melting transitions. The
freezing and melting packing fractions are respectively
F
0.494 and
M
0.545, Hoover and Ree (1968). Fuller (1975) derives the expression
C
n
= 10 n
2
+2 for a perfect fcc lattice. C
1
for a perfect fcc lattice is 12, since
the unit cell is the rhombic dodecahedron. However, due to a topological in-
stability the average number of faces of a Voronoi polyhedron in a perturbed
fcc lattice is 14, as shown by Troadec et al. (1998). We have outlined the orig-
inal proof in the appendix of Chapter 3. Thus, the thermodynamic structure
C
1
approaches 14 instead of 12.
As in the sheared inelastic hard-disk structures, the near-elastic sheared uid
structure C
n
are identical to the thermodynamic structure C
n
.
When the sheared structures undergo the shear ordering transition, the C
n
drops close to the thermodynamic solid C
n
value. On shear ordering, close-
packed two-dimensional layers slide past each other. The shear ordering tran-
sition causes shear thinning in colloids and dense suspensions, since layers
sliding past each other offer lesser resistance to shear than a disordered struc-
ture, Woodcock (1985). The sheared hard-sphere C
n
show negligible system
size dependence, as observed from Table 4.2.
The near-elastic ( = 0.999) C
n
drop to the thermodynamic solid phase val-
ues at a higher packing fraction than the thermodynamic uid branch, due
to the suppression of crystal nucleation in a homogeneous shear eld, Blaak
et al. (2004). For = 0.999, 0.99, 0.95, 0.90 and 0.80, the shear-ordering
Chapter 4 91
0 0.2 0.4 0.6
14
14.2
14.4
14.6
14.8
15
15.2
15.4
15.6
0.45 0.5 0.55
14
14.2
14.4
14.6
14.8
PSfrag replacements
C
1
Figure 4.7: C
1
for inelastic hard-spheres sheared in the preferred orientation, at = 0.70
(), 0.80 (),0.90 (), 0.95 (+), 0.99 () and 0.999 (), compared with thermodynamic
structures (). Thermodynamic data averaged for 1000 con gurations of 256 hard-spheres,
with PBC. Sheared structure data averaged for 675 con gurations of 384 hard-spheres (6
stacks of 8 8 hard-spheres), with PBC.
packing fraction varies within the narrow range of 0.52-0.53. However, for
= 0.70 the shear-ordering packing fraction is between 0.54 and 0.55 (in-
set of Fig. 4.7). Hence, we hold that, the shear-ordering packing fraction
increases with increasing inelasticity, as in the hard-disk system.
The shear-ordered C
n
are lower than those of the thermodynamic solid, ev-
ident clearly in the C
2
data. From Table 4.3 we note the following: C
n
of
the fcc and the hexagonal close-packed (hcp) lattices are nearly identical and
the sheared structure C
n
are inbetween those of the bcc and fcc/hcp lattices,
at the same packing fraction. This is indicative of the coexistence of a bcc-
like order along with the fcc order in the dense sheared inelastic hard-spheres
structures. We discuss this further while analyzing P
n
below.
0 0.2 0.4 0.6
55
60
65
70
PSfrag replacements
C
2
Figure 4.8: C
2
for inelastic hard-spheres sheared in the preferred orientation, at = 0.70
(), 0.80 (),0.90 (), 0.95 (+), 0.99 () and 0.999 (), compared with thermodynamic
structures (). Averaging as in Fig. 4.7.
2 1 0 1 2
2
1.5
1
0.5
0
0.5
1
1.5
2
PSfrag replacements
y/
z
/
2 1 0 1 2
2
1.5
1
0.5
0
0.5
1
1.5
2
PSfrag replacements
y/
z/
x/
z
/
Figure 4.9: The pair-distribution function projected on the velocity gradient-vorticity (yz)
plane (left) and the velocity-vorticity (xz) plane (right), for the preferred orientation shear
of inelastic hard-spheres at = 0.54 and = 0.99. Averaged for 5000 con gurations of
384 hard-spheres, with PBC.
Chapter 4 93
Table 4.2: System size dependence check C
1
and C
2
for inelastic hard-spheres sheared in the
preferred orientation, for a constant coef cient of restitution = 0.70.
Run C
1
C
2
0.45 I
a
14.8317 59.4936
II
b
14.8335 59.5084
III
c
14.8263 59.4719
IV
d
14.8270 59.4691
0.50 I 14.7323 58.5598
II 14.7301 58.5493
III 14.7371 58.6100
IV 14.7235 58.5273
0.57 I 14.0701 52.7385
II 14.0705 52.7770
III 14.0942 52.8972
IV 14.1058 53.0466
0.61 I 14.0393 52.4071
II 14.0336 52.4847
III 14.0521 52.4887
IV 14.0594 52.5013
a
Averaged for 1200 con gurations of 216 (6 stacks of 6 6) hard-spheres,
l
x
: l
y
: l
z
= 6 : 6
2/3 : 6
3/2.
b
l
x
: l
y
: l
z
= 8 : 6
2/3 : 8
3/2.
c
l
x
: l
y
: l
z
= 8 : 9
2/3 : 8
3/2.
d
l
x
: l
y
: l
z
= 10 : 9
2/3 : 10
3/2.
Table 4.3: C
1
and C
2
for different structured systems at = 0.59.
System C
1
C
2
bcc
a
14.0231 50.2857
hcp
b
14.0687 53.7314
fcc
c
14.0695 53.6867
sheared
d
14.0456 52.4773
a
Averaged for 1000 con gurations of 432 hard-spheres, in a cubic box.
b
l
x
: l
y
: l
z
= 8 : 6
2/3 : 8
3/2.
c
d
= 0.90, averaged for 675 con gurations of 384 hard-spheres,
l
x
: l
y
: l
z
= 8 : 6
2/3 : 8
3/2.
During the plastic deformation of fcc solids (ABC stacking), a close-packed
(111) plane slides over another in a zig-zag path, alternating between type-B and
type-C sites, Cottrell (1953). This phenomenon is noted also in colloidal suspen-
sions, Ackerson (1990). This motion corresponds to the zig-zag haze in the pair
distribution function projected on the velocity-vorticity (xz) plane, Fig. 4.9.
In Fig. 4.10, we present the angular-averaged pair distribution function at con-
tact, g(). For comparison we show that g() values for the thermodynamic uid
and solid, computed from the equations of states as
g() =
Z 1
B
2

, (4.8)
where Z = pv/(k
B
T) is the compressibility factor and B
2
= (2/3)
3
is the hard-
sphere second virial coefcient. For the thermodynamic uid phase we use the
Carnahan-Starling equation (Carnahan and Starling (1970)),
Z =
1 + +
2
3
(1 )
3
. (4.9)
For the thermodynamic solid phase we use the equation of Young and Alder (1979),
Z =
3
a
+ 2.566 + 0.55 a 1.19 a
2
+ 5.95 a
3
, (4.10)
Chapter 4 95
where a = (1 y)/y is the fractional free-volume and y = /
c
is the normalized
packing fraction and
c
= /(3
2) is the regular close-packed packing fraction.

Some of the conclusions which we drew from the analysis of C
n
, can indeed be
drawn from g() also, for example, the shear ordering transitions and the approach
to thermodynamic structures by the near-elastic sheared structures. However, a
structural analysis based on g() could be misleading, for example, in Fig. 4.10 the
highly inelastic solid structures seem to be a continuation of the thermodynamic
uid phase. But from our analysis of C
1
in Fig. 4.7, we know that such an inference
is erroneous. From Voronoi partitioning (Section 3.2), note that g() = g
1
() is
composed of C
1
and f
1
(). As packing fraction increases C
1
decreases (except
during amorphization, as noted below) and f
1
() increases. And both C
1
and f
1
()
decrease suddenly across an ordering transition. Since g() is composed of terms
with such mixed trends, structural inferences based on it could be misleading, as
shown above. Hence, the C
n
are superior structural indices than g().
The P
n
for thermodynamic and sheared structures are given in Figs. 3.12, 3.13,
4.11 and 4.12. Due to the topological instability of the fcc lattice, near the regular
close packed limit, the thermodynamic solid has Voronoi polyhedra with faces rang-
ing from 12 to 18 (as observed in Figs. 3.12 and 3.13) with a mean at 14, the original
proof is outlined in the appendix of Chapter 3. Comparing the P
n
of thermody-
namic and sheared structures, we see that the 14-faceted polyhedron incidence has
increased in the sheared solid phase, at the expense of the 12, 13, 15 to 18 faceted
polyhedra. Due to the depletion of the high-faceted polyhedra, C
1
=
nP
n
of the
sheared solid structure is lower than that of the thermodynamic solid structure. This
effect gets amplied for the second nearest neighbors and hence more prominent in
Fig. 4.8.
The Voronoi polyhedron of the bcc lattice is the truncated octahedron, which
has six square and eight hexagonal faces, a total of 14 faces. The bcc polyhedron
is topological stable to perturbations, Tanemura et al. (1977); Hsu and Rahman
(1979), unlike the fcc or hcp polyhedra, Troadec et al. (1998). From Table 4.4 we
note the following: P
n
of the fcc and hcp lattices are nearly identical and the sheared
structure P
n
are inbetween those of the bcc and fcc/hcp lattices, at the same packing
fraction. This high incidence of 14-faceted polyhedron is a clear indication of the
0.3 0.4 0.5 0.6
5
10
15
20
25
30
PSfrag replacements
g
(
)
Figure 4.10: g() for the preferred orientation shear of inelastic hard-spheres, at = 0.60
(), 0.70 (), 0.80 (), 0.90 (), 0.95 (+), 0.99 (), compared with Carnahan-Starling
equation for the thermodynamic uid phase (continuous line) and the Young-Alder equation
for the solid phase (dashed line). Data sets averaged for 675 con gurations of 384 hard-
spheres, with PBC.
0 0.2 0.4 0.6
10
2
10
1
10
0
PSfrag replacements
P
n
Figure 4.11: P
n
for sheared inelastic hard-sphere structures at = 0.90, for n =12 (), 13
(), 14 (), and 15 (). Averaging as in Fig. 4.7.
Chapter 4 97
0 0.2 0.4 0.6
10
5
10
4
10
3
10
2
10
1
10
0
PSfrag replacements
P
n
Figure 4.12: P
n
for sheared inelastic hard-sphere structures at = 0.90, for n =11 (), 16
(), 17 (), 18 (), and 19 (+). Averaging as in Fig. 4.7.
Table 4.4: Percentage distribution of the number of faces of the Voronoi polyhedra for different
structured systems at = 0.59.
P
n
100
n bcc
a
hcp
b
fcc
c
sheared
d
12 0.01 4.03 3.44 0.68
13 1.89 23.65 23.44 15.96
14 93.95 40.76 41.83 63.56
15 4.07 25.07 25.66 17.80
16 0.07 5.95 5.28 1.93
17 0 0.51 0.35 0.07
18 0 0.01 0.01 0
a
b
l
x
: l
y
: l
z
= 8 : 6
2/3 : 8
3/2.
c
d
= 0.90, averaged for 675 con gurations of 384 hard-spheres,
l
x
: l
y
: l
z
= 8 : 6
2/3 : 8
3/2.
0 0.2 0.4 0.6
14
14.2
14.4
14.6
14.8
15
15.2
15.4
15.6
PSfrag replacements
C
1
Figure 4.13: C
1
for sheared inelastic hard-sphere structures at = 0.99, at the preferred ()
and an unpreferred () orientation, compared with thermodynamic structures (). Averag-
ing as in Fig. 4.7. The experimental observations by Finney (1970) at
DRP
0.64 are
().
presence of a bcc-like structure in dense sheared inelastic hard-spheres. During
martensitic transformations, a rapidly quenched fcc metallic structure gets sheared
to the bct structure, by a mechanism called the Bain distortion, Nishiyama (1978).
When a cubic lattice (with lattice parameters a = b = c) is compressed or stretched
in one direction it becomes a tetragonal lattice (with lattice parameters a = b ,= c).
The bct structures similarly formed in the dense homogeneously sheared inelastic
hard-spheres leave a bcc-like signature in the C
n
and P
n
data. A detailed bond-
orientational analysis of dense sheared microstructures is presented in Chapter 5.
Now we consider shear in an unpreferred orientation, with the (001) plane paral-
lel to the velocity-vorticity (xz) plane. Since, crystalline solids are anisotropic, their
properties vary with direction. Woodcock (1985) is critical about the unpreferred
orientation shear simulations, and considers the observations as artifacts. However,
we consider such simulations as indicative of metastable or transient states, since
Chapter 4 99
shearing in an unpreferred orientation the structure melts and (given sufcient time)
recrystallizes in the preferred orientation. In Fig. 4.13 we compare the C
1
for the
preferred and an unpreferred orientation, from which we observe the following:
For uid states, the orientation of the initial conguration does not matter,
hence the structures are identical for both the preferred and unpreferred ori-
entations.
When the system sheared in an unpreferred orientation, the lattice melts and
recrystallizes in the preferred orientation, hence
u
O

p
O
, where
u
O
and
p
O
are the shear ordering packing fractions in an unpreferred and the preferred
orientations. In Fig. 4.13,
u
O
0.59 and
p
O
0.53. In Fig. 4.14, for
= 0.60 note that the lattice orientation has not stabilized, compare with the
lattice orientation in Fig. 4.9.
In an unpreferred orientation shear, at packing fractions above shear ordering
packing fraction, the system gets disordered or amorphized, as observed in
Fig. 4.14. This phenomenon is analogous to the strain-induced amorphiza-
tion of nanocrystals, Ikeda et al. (1999) (in this work also shear is along the
(001) plane). This is a metastable state, because on annealing or given in-
nite time the sample will recrystallize. While amorphization was absent in
the preferred orientation shear, it is observed in an unpreferred orientation
shear. We reason its cause as follows: X-ray diffraction studies show that
plastic deformation does not disorder a crystal, even though the defect con-
centration increases. Most of the plastic deformation is by slipping, in which
the particle displacements are along the slip planes. Hence, we infer that if
the imposed displacement or velocity lies in a slip plane, the lattice need not
break. This corroborates with the fact that amorphization was not observed
in the preferred orientation shear (more below). Now, when the imposed dis-
placement or velocity does not lie in a slip plane, it is intuitively obvious that
simultaneous shear and slip can break the lattice. We consider this to be a
cause for the metastable amorphization in an unpreferred orientation shear. If
the relaxation time of the system is very small as compared to the shear, this
metastable amorphization will not be observed, since the system will recrys-
tallize quickly in the preferred orientation. In most colloidal system, at higher
shear rates shear thickening is observed. A steady state amorphization is con-
sidered to be its cause, Hoffman (1972, 1998), and it is observed in Stokesian
dynamics simulations, Sierou and Brady (2002). Shear thickening by amor-
phization requires that, at a given packing fraction, the structure change with
shear rate. Hence, shear-thickening cannot be observed with the constant
COR model, for which the structures are shear rate independent. Below we
demonstrate that using a velocity-dependent COR steady state amorphization
can be realized even in the preferred orientation.
We speculate that for packing fractions well above the dense random pack-
ing, an unpreferred orientation shear will break the lattice into polycrystalline
structure and over long times the grains will reorient and crystallize in the
preferred orientation. Our simulation sizes and times are too small to observe
such a phenomenon.
From Fig. 4.13 we note that the C
1
for amorphized structures are close to that
reported by Finney (1970) for two different experimental realizations of dense
random packing, C
1
= 14.2510.015 and 14.280.05. Fig. 4.15 shows that
the distribution of the faces of the Voronoi polyhedra (P
n
) of such structures
match reasonably with those of the swelled random structures, except for the
higher incidence of some high-faceted polyhedra at the expense of the low-
faceted ones. Structural relaxation using the swelling algorithm can rectify
the same. Thus, the Lees-Edwards boundary condition for nearly elastic hard-
spheres at 0.64, in an unpreferred orientation, followed by a few swelling
cycles for structural relaxation, offers a nearly deterministic route to generate
dense random packings.
In an unpreferred orientation shear, we nd that the shear ordering fraction
increases with inelasticity in a wider range than in the preferred orientation,
data not shown.
Before we present the simulation results, we demonstrate the possibility of
achieving of steady state amorphization in the preferred orientation with the velocity-
dependent COR. From Fig. 4.7, note that shear ordering occurs in the range =
0.52 0.53 for = 0.999 0.8 and at 0.55 for = 0.70. Suppose the
Chapter 4 101
2 1 0 1 2
2
1.5
1
0.5
0
0.5
1
1.5
2
PSfrag replacements
y/
z
/
2 1 0 1 2
2
1.5
1
0.5
0
0.5
1
1.5
2
PSfrag replacements
y/
z/
y/
z
/
Figure 4.14: The pair-distribution function projected on the velocity gradient-vorticity (yz)
plane for an unpreferred orientation shear of inelastic hard-spheres at = 0.60 (left) and
= 0.63 (right), with = 0.99. Averaged for 5000 con gurations of 384 hard-spheres,
with PBC.
10 12 14 16 18 20 22 24
10
6
10
4
10
2
10
0
PSfrag replacements
n
P
n
Figure 4.15: P
n
for inelastic hard-spheres sheared at an unpreferred orientation, at =0.63
and = 0.99 for different system sizes (N= 256 (), 0.500 (), 864 () and 1372 (+)),
compared with the swelled random structures ().
the velocity-dependent COR sheared structures could be related (at least approxi-
mately) to that at some constant COR. At = 0.545, say the shear rate is initially
such that the average COR is greater than 0.80, then the system is shear ordered.
Now, if the shear rate is increased such that the average COR decreases below 0.70,
then the system will get amorphized. Even if the system is globally in preferred ori-
entation, near defects like edge dislocations or grain boundaries, a < 111 > plane
may be locally mis-oriented and not contain the velocity. Such locations act as
nuclei for disorder, since simultaneous shear and slip on such planes will generate
more defects. When the local amorphization rate exceeds the local crystallization
rate, a steady amorphized state results. In our simulations, at state points just before
amorphization, we found more disorder near the top and bottom boundaries of the
simulation cell. Probably the top and bottom boundaries of the simulation cell be-
have like grain boundaries, due to velocity differences between the top, central and
bottom boxes.
Ackerson et al. (1986) have observed shear thickening in suspension of polystyrene
latex spheres (0.109 m) suspended in density-matched water/heavy-water mixture
at 400 s
1
shear rate. Cao and Ahmadi (1995) use a constant COR = 0.90 for
polystyrene spheres. Since we could not nd the velocity-dependent COR data for
polystyrene in the literature, we use that of Nylon reported by Labous et al. (1997),
the average COR is = 0.97 0.03 for spheres in the range 6.35 mm to 25.4
mm, and v
Y
= 9 m/s. Their data ts our COR model Eq. 1.5, with B = 0.12
and C = 0.44. With these material constants, a steady state amorphization is ob-
served in the preferred orientation shear, for 20 mm diameter spheres near a critical
shear rate
c
2500 s
1
, as seen from the projected pair distribution function in
Fig. 4.16.
From the data in Table 4.5, we observe that
c
increases with decreasing par-
ticle diameter. For shear-thickening dense suspensions
c

2
, Barnes (1989);
Hoffman (1998). However, in our simulations we observe that
c

1
, i.e.
c

constant ( = 10 mm,
c
4900 5000 s
1
; = 15 mm,
c
3500 3600 s
1
;
= 20 mm,
c
2400 2500 s
1
; = 25 mm,
c
1900 2000 s
1
). The
inverse linear proportionality of
c
with is predicted for the shear thickening of
granular suspensions by Savage and Jeffrey (1980) (quoted in Barnes (1989)). In
Chapter 4 103
2 1 0 1 2
2
1.5
1
0.5
0
0.5
1
1.5
2
PSfrag replacements
x/
z
/
2 1 0 1 2
2
1.5
1
0.5
0
0.5
1
1.5
2
PSfrag replacements
x/
z/
x/
z
/
Figure 4.16: The pair-distribution function projected on the velocity-vorticity (xz) plane for
the preferred orientation shear of 20 mm diameter spheres at shear rates = 2400 s
1
(left)
and 2500 s
1
(right), at = 0.545 for the velocity dependent COR model. Averaged for
5000 con gurations of 384 hard-spheres, with PBC.
our system, at a given packing fraction, v
Y
/(
c
) is the only material dimensionless
group, and insisting its constancy at amorphization, the above scaling is recovered.
Note that the average COR of Nylon is higher than that of polystyrene, hence we re-
quire higher shear rates to observe amorphization. We have used sphere diameters
for much higher than that of polystyrene (0.109 m), in Table 4.5, so that amor-
phization in observed in the typical range of shear rates (0 to 10
5
s
1
). However,
note that in colloidal suspension due to the screened electrostatic interactions (mod-
eled with Yukawa potential, Stevens and Robbins (1993)), the effective hard-sphere
diameter is larger than the actual sphere diameter.
4.5 Conclusions
Inelastic hard-particles are the simplest model for rapid granular matter. We have
generated the homogeneously sheared inelastic hard-disk and hard-sphere structures
using the Lees-Edwards boundary condition, and performed a Voronoi neighbor
analysis for these structures. In most part of the chapter, we have used the constant
coefcient of restitution model for inelastic collisions. Due to the lack of an intrinsic
Table 4.5: Shear amorphization with velocity-dependent COR in the preferred orientation, at
= 0.545 at different sphere diameters. Averaged for 675 con gurations of 384 hard-spheres
(6 stacks of 8 8 hard-spheres), l
x
: l
y
: l
z
= 8 : 6
_
2/3 : 8
3/2.
mm s
1
C
1
C
2
10 4800 14.0959 53.1667
4900 14.0978 53.1682
5000 14.5478 57.1050
5200 14.5705 57.2308
15 3400 14.0984 53.1841
3500 14.0953 53.1508
3600 14.5504 57.0752
3700 14.5674 57.1976
20 2300 14.1024 53.1617
2400 14.0959 53.1667
2500 14.5592 57.2328
3000 14.5627 57.1953
25 1800 14.0974 53.1820
1900 14.0950 53.1631
2000 14.5478 57.1050
2200 14.5508 57.1120
Chapter 4 105
energy scale, the resultant structures are shear rate independent. Hence, to study
the shear-dependent structures, we have used a velocity-dependent coefcient of
restitution. We propose a newfunctional form for the velocity-dependent coefcient
of restitution, Eq. 1.5, which accommodates both the viscoelastic and the plastic
deformation scalings.
In a given conguration, all the spheres surrounding a central sphere are parti-
tioned layer-wise, and characterized by nth neighbor coordination number (C
n
) and
nth neighbor position distribution function [f
n
(r)]. We have shown that the total
information contained in the pair distribution function, g(r), can be partitioned into
these sets of neighbor statistics. The distribution of the number of Voronoi bounding
surfaces (P
n
) is also of interest, because C
1
is its mean. These neighbor statistics
are sensitive microstructural changes like thermodynamic freezing, shear ordering
and shear amorphization transitions, and can distinguish thermodynamic, quenched
and sheared structures.
Both the sheared inelastic hard-disk and hard-sphere (in the preferred orien-
tation) structures show the following: In the near-elastic limit these sheared uid
statistics approach that of the thermodynamic uid values. When these sheared
neighbor statistics approach the thermodynamic solid phase values, shear ordering
transition takes place. On shear ordering, close-packed layers of particles slide over
each other. In colloids and dense suspensions, such a layer-wise ow causes shear-
thinning, since sliding close-packed layers offer lesser resistance to shear than a
disordered structure. The Voronoi statistics clearly show the suppression of crys-
tal nucleation by homogeneous shear. As inelasticity increases the shear ordering
packing fraction increases. Due to a topological instability, a thermodynamic hard-
sphere solid, has polyhedra with faces 12 to 18, with the mean at 14. In shear
ordered inelastic hard-sphere structures there is a high incidence of 14-faceted poly-
hedra and a consequent depletion of polyhedra with faces 12, 13, 15 to 18, due to
the formation of bct structures. These bct structures leave a bcc-like signature in the
C
n
and P
n
data.
In an unpreferred orientation shear, at packing fractions above shear ordering,
the structures get amorphized. This is a metastable amorphization, since the system
will eventually recrystallize in the preferred orientation. When the imposed dis-
placement or velocity does not lie on a slip plane, simultaneous slip and shear can
break the lattice. We consider this as a cause for the metastable amorphization ob-
served in an unpreferred orientation shear. These structures are nearly identical to
the dense random structures, except for the higher incidence of some high-faceted
polyhedra at the expense of some low-faceted ones. This aberration can be rectied
by a structural relaxation using the swelled random algorithm. Thus, the Lees-
Edwards boundary condition simulation for nearly elastic hard-spheres sheared in
an unpreferred orientation, followed by a structural relaxation using the swelled
random algorithm, offers a nearly deterministic algorithm to generate dense ran-
dom packings.
A steady state amorphization is the cause of shear thickening observed in dense
suspensions. The constant coefcient of restitution sheared structures are shear
rate independent, hence they cannot amorphize as shear rate is varied. We have
demonstrated a steady state amorphization with a velocity-dependent coefcient of
restitution. Even when the system is globally in the preferred orientation, the slip
planes are locally mis-oriented near edge dislocations and grain boundaries. At
these locations, shear induces local amorphization, since the velocity does not lie
on these local slip planes. When the local amorphization rate exceeds the local
recrystallization rate, a steady amorphized state results. We observe that the critical
shear rate for amorphization is inversely proportional to the sphere diameter, as
predicted by Savage and Jeffrey (1980) for granular suspensions, unlike the inverse
square scaling observed in the colloidal suspensions.
Chapter 5
Bond-orientational analysis of
hard-disk and hard-sphere structures
We report the bond-orientational analysis results for the thermodynamic, random
and homogeneously sheared inelastic hard-disk and hard-sphere structures. The
thermodynamic structures show a sharp rise in the order across the freezing tran-
sition. The random structures show the absence of crystallization. However, in
random hard-disk congurations local crystallization is high due to the lack of ge-
ometric frustration in two-dimensions. Due to the suppression of crystal nucleation
by the homogeneous shear, the sheared structures get ordered at a packing frac-
tion higher than the thermodynamic freezing packing fraction. On shear ordering,
strings of close-packed hard-disks in two-dimensions and close-packed planes of
hard-spheres in three-dimensions, oriented along the velocity direction, slide past
each other. Such a ow creates a considerable amount of four-fold order in two-
dimensions and body-centered tetragonal (bct) structure in three-dimensions. These
transitions are the ow analogues of the martensitic transformations occurring in
metals due to the stresses induced by a rapid quench. While the transition in the
metallic systems are due to diffusionless lattice distortions occurring at rates com-
parable to the speed of sound, the transition in the owsystems are due to the sliding
of close-packed layers past each other occurring over the ow time scales. In hard-
disk structures, using the bond-orientational analysis we show the presence of four-
fold order. When close-packed spheres slide past each other, in addition to the bct
107
108 Bond-orientational analysis of hard-disk and hard-sphere structures
structure, the hexagonal close-packed (hcp) structure is formed due to the random
stacking faults. In sheared inelastic hard-sphere structures, even though the global
bond-orientational analysis shows that the system is highly ordered, a third-order
rotational invariant analysis shows that only about 40% of the spheres have face-
centered cubic (fcc) order, even in the dense and near-elastic limits, clearly indicat-
ing the coexistence of multiple crystalline orders. Using the Honeycutt-Andersen
pair analysis and an analysis based on the 14-faceted polyhedra having six quadri-
lateral and eight hexagonal faces, we show the presence of bct and hcp signatures
in shear ordered inelastic hard-spheres. Thus, our analysis shows that the dense
sheared inelastic hard-spheres have a mixture of fcc, bct and hcp structures. These
results are reported in Kumar and Kumaran (2006a).
5.1 Introduction
In this chapter we do the bond-orientational analysis for thermodynamic (NVE-
MC), swelled random and sheared inelastic hard-disk and hard-sphere structures,
with particular focus on the dense sheared inelastic hard-particle microstructure.
Every particle is said to be connected with its Voronoi neighbor by a bond. The ori-
entation of these bonds are characterized by angular measures, sines/cosines in two-
dimensions and spherical harmonics in three-dimensions. The rotationally invariant
functions dened on these angular measures are used in the bond-orientational anal-
ysis to describe the angular order of the local neighborhood, Halperin and Nelson
(1978); Steinhardt et al. (1983). Such an analysis is useful in characterizing the
microstructures and quantifying the degree of local crystallization.
In Section 5.2 we present the global and local bond-orientational analysis of
hard-disk structures. The six-fold and four-fold orders in the hard-disk systems are
monitored at different packing fractions. The thermodynamic hard-disk structures
have negligible four-fold order, and the six-fold order increases sharply across the
freezing transition. The random hard-disk structures have a low global six-fold
ordering, but there is a considerable amount of local crystallization, since the tri-
angular structure maximizes both the local and global packing. This phenomena,
called the lack of geometric frustration, is considered to be the cause for the insta-
Chapter 5 109
bility of dense random hard-disk packings under compressive forces, Quickenden
and Tan (1974); Hinrichsen et al. (1990). We study the effect of the success rate
of the trial displacements used in the swelling algorithm on the local crystallization
observed in the resultant random structures. At lower success rates, the large trial
displacements tend to equilibrate the nonequilibrium structures and hence the local
crystallization is high. At higher success rates, the trial displacements are small and
the swelling process tends to lock the particles into random structures. In a homoge-
neous shear eld, it is known that crystal nucleation gets suppressed. This phenom-
ena is, for example, observed in the Brownian dynamics simulations of colloidal
suspension by Butler and Harrowell (1995); Blaak et al. (2004) and in the results
of Chapter 4. Due to the suppression of crystal nucleation, the sheared structures
get ordered at a packing fraction higher than the thermodynamic freezing packing
fraction. On shear ordering, strings of close-packed hard-disks, oriented along the
velocity direction, slide past each other. This phenomena has been demonstrated by
Stancik et al. (2003) in the shear of a monolayer of polystyrene beads suspended
at the decane-water interface, also noted in Chapter 4. When close-packed strings
of hard-disks slide past each other, considerable amount of four-fold order is gener-
ated. This triangle-to-square transition has been observed in the shear of a conned
monolayer of colloidal suspensions by Weiss et al. (1995). The symmetry change
involved is the two-dimensional analogoue of the martensitic transformations ob-
served in metals due to stresses induced by a rapid quench. While the transition
in the metallic systems are due to diffusionless lattice distortions occurring at rates
comparable to the speed of sound, Callister Jr. (2000), the transition in the ow sys-
tems are due to the sliding of close-packed layers past each other occurring over
ow time scales.
In Section 5.3 we present the bond-orientational analysis results for the hard-
sphere systems. The global analysis shows the onset of order in thermodynamic
structures across the freezing transition, the absence of order in random structures
and the suppression of nucleation in the homogeneously sheared structures. In
sheared structures, even though the global analysis shows considerable level of or-
der, the local analysis using a third-order rotational invariant, devised by Mitus et al.
(1995), shows that the fraction of spheres having face-centered cubic (fcc) order is
only about 40%, even in the dense and near-elastic limits. This clearly indicates the
coexistence of multiple crystalline orders, which is explained as follows. On shear
ordering, close-packed planes of spheres slide past each other in a zig-zag path,
with the close-packed planes oriented parallel to the velocity-vorticity plane. This
phenomena is observed in the plastic deformation of fcc crystals (Cottrell (1953)),
colloidal suspensions (Ackerson (1990)), and our results in Chapter 4. When the
close-packed layers slide thus past each other, the spheres in a layer tend to occupy
a position vertically above the centers of the spheres in the adjacent layers. This
leads to the formation of bct structures. This fcc-to-bct transition is observed in the
martensitic transformations exhibited by rapidly quenched metals. Note that the fcc
and the hexagonal close packing (hcp) differ only in the stacking order, ABC and
ABA respectively. When close-packed planes slide past each other, the random
stacking faults form hcp structures. This phenomena is observed in the shear of
colloidal suspensions (Loose and Ackerson (1994)) and polymer solutions (Bang
and Lodge (2003)). Using the pair analysis of Honeycutt and Andersen (1987) and
an analysis based on the 14-faceted polyhedra having six quadrilateral and eight
hexagonal faces, we show the presence of hcp and bct signatures in shear ordered
inelastic hard-spheres. Thus, our analysis shows that the dense sheared inelastic
hard-spheres have a mixture of fcc, hcp and bct structures.
5.2 Hard-disk structures
The mfold global bond-orientational order parameter, Halperin and Nelson (1978),
is dened as
m
= exp (i m )), (5.1)
where ) is the average over all the Voronoi neighbor bonds in the system and
is the angle formed by a bond with respect to some arbitrary axis. For a per-
fect mfold ordered system, [
m
[ is unity, and lower otherwise. To recognize the
freezing of hard-disks into hexagonal close-packed structure [
6
[ is used, Mazenko
(2000). Fig. 5.1 shows the [
6
[ for thermodynamic, random and sheared inelastic
hard-disk structures, from which we observe the following:
At the freezing packing fraction (
F
0.691, Alder and Wainwright (1962)),
[
6
[ for the thermodynamic hard-disk structure rises sharply due to the onset
Chapter 5 111
0.4 0.5 0.6 0.7 0.8 0.9
0
0.2
0.4
0.6
0.8
1
PSfrag replacements
6
[
Figure 5.1: |
6
| for hard-disks: thermodynamic structures (), random structures () gen-
erated at 90% success rate, and sheared inelastic structures at = 0.80 (), 0.90 (), 0.95
(+) and 0.99 (). Thermodynamic and sheared structures data averaged for 10000 con g-
urations, random structures data averaged for 1000 con gurations, of 256 hard-disks, with
periodic boundary conditions (PBC).
of hexagonal order. Above the melting packing fraction (
M
0.716, Alder
and Wainwright (1962)) the [
6
[ is close to unity indicating highly correlated
local order.
[
6
[ for the dense random hard-disk structures is low, due to the absence of
correlated order. In hard-disk packings, the structure maximizing the local
and the global packing are identical, i.e. the triangular or the hexagonal struc-
ture. This lack of geometric frustration is considered to be the cause of the
instability of dense random hard-disk packings against compressive forces,
Quickenden and Tan (1974); Hinrichsen et al. (1990). (In three dimensions,
the locally densest packing viz the tetrahedral packing is not space-lling.
This geometric frustration is considered to be the cause of the stability of
dense random hard-sphere packings against compressive forces.) Thus, there
is a considerable degree of local crystallization in dense random hard-disk
packings, shown below by the local bond-orientational analysis. However,
the orientation of the local crystalline domains are uncorrelated. Hence,
6
has a low value for the dense random hard-disk structures.
Even for = 0.99, i.e. near-elastic limit, the sheared structures get ordered
only above 0.77. The shear ordering packing fraction being higher than
the thermodynamic freezing packing fraction shows the suppression of crys-
tal nucleation by homogeneous shear. This phenomena is also observed in the
Brownian dynamics simulations of colloidal suspension by Butler and Har-
rowell (1995); Blaak et al. (2004), and in Chapter 4.
For = 0.99, 0.95 and 0.90, the shear ordering packing fractions are 0.77,
0.81 and 0.83, respectively. This shows that, as inelasticity increases shear
ordering occurs at a higher packing fraction.
On shear ordering, strings of close-packed hard-disks, oriented in the velocity
direction, slide past each other. This phenomena was demonstrated in the shear
of a monolayers of polystyrene spheres suspended at the decane-water interface
by Stancik et al. (2003). When strings of close-packed hard-disks to slide past
each other, the distance between adjacent strings need to be greater than , at least
locally, while the distance between adjacent strings is (
3/2) in the hexagonal

Chapter 5 113
0.4 0.5 0.6 0.7 0.8 0.9
0
0.05
0.1
0.15
PSfrag replacements
8
[
Figure 5.2: |
8
| for hard-disks: thermodynamic structures (), random structures () gen-
erated at 90% success rate, and sheared inelastic structures at = 0.80 (), 0.90 (), 0.95
(+) and 0.99 (). Averaging as in Fig. 5.1.
close-packed structure. When strings of close-packed hard-disks slide past each
other, square or rhombic structures are formed. Weiss et al. (1995) have observed
this triangle-to-square transition in the ow of conned monolayer colloidal sus-
pensions, and consider it the two-dimensional ow analogue of martensitic trans-
formations observed in metallic systems occurring due to the stresses induced by a
rapid quench. Due to the topological instability of the square lattice to slight per-
turbations (noted in Section 3.5), [
8
[ is commonly used to monitor the distorted
four-fold order instead of [
4
[. The [
8
[ data for the thermodynamic, random and
sheared structures are given in Fig. 5.2, from which we observe the presence of a
signicant amount of four-fold order in the sheared hard-disk structures.
Next we proceed to the local analysis, using the method of Weiss et al. (1995).
The mfold local bond-orientational order parameter of disk j with N
1
Voronoi
neighbors is
m
=
1
N
1
N
1
k=1
exp (i m
k
). (5.2)
0 2 4 6 8
0
0.2
0.4
0.6
0.8
1
PSfrag replacements
t
f
6
,
f
8
Figure 5.3: Evolution of f
6
(continuous line) and f
8
(dotted line) in a sheared inelastic
hard-disk simulation run at = 0.72 and = 0.90. Initial con guration was the hexagonal
lattice.
If [
6
[ > [
8
[ then a disk is said to have a six-fold coordination, and four-fold
coordination otherwise. The fraction of disks having six-fold and four-fold coor-
dinations are f
6
and f
8
(= 1 f
6
) respectively. When the homogeneous shear is
switched on at time t = 0 in an equilibrated hard-disk system, above its shear or-
dering packing fraction, due to the sliding of close-packed hard-disk strings, the
six-fold coordination decreases and the four-fold coordination increases, and until
the system reaches a steady state, Fig. 5.3. If the initial packing is disordered, even
though the system packing fraction is above the shear ordering packing fraction,
then the system initially gets ordered into layers, before the layers start sliding past
each other. Then, during the initial ordering f
6
increases, before it starts decreasing
due to the sliding motion. The symmetry change involved in this triangle-to-square
transition is analogous to the martensitic transformation observed in metallic sys-
tems. However, there are differences in the underlying mechanism. The martensitic
transformation occurs by a diffusionless lattice distortion, at rates comparable to the
speed of sound, Callister Jr. (2000). The transition observed in these dense ow sys-
Chapter 5 115
0.4 0.5 0.6 0.7 0.8 0.9
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0.9
1
PSfrag replacements
s
6
Figure 5.4: s
6
for hard-disks: thermodynamic structures (), random structures generated at
success rates 10% (), 30% (), 50% (+), 70% () and 90% (). Averaging as in Fig. 5.1.
tems are due to the formation, unlocking and sliding of close-packed layers, which
occurs over ow time scales.
A disk is considered to be ordered, if [
6
[ > 0.5 or [
8
[ > 0.5. The fraction of
ordered six-fold and four-fold disks are s
6
and s
8
respectively. The s
6
and s
8
data
for the thermodynamic and random structures are given in Figs. 5.4 and 5.5, from
For >
F
, most of the disks in the thermodynamic structure are six-fold
ordered disks.
For <
F
, the random structures are identical with the thermodynamic
structures, while they are distinct for >
F
.
In random hard-disk structures, due to the lack of geometric frustration, there
is a signicant amount of crystallites. However, this was not evident in the
global analysis (Fig. 5.1) since the orientation of the local crystallites were
uncorrelated.
0.4 0.5 0.6 0.7 0.8 0.9
0
0.05
0.1
0.15
0.2
0.25
0.3
PSfrag replacements
s
8
Figure 5.5: s
8
for hard-disks: thermodynamic structures (), random structures generated at
success rates 10% (), 30% (), 50% (+), 70% () and 90% (). Averaging as in Fig. 5.1.
At lower success rates the fraction of crystallites (s
6
) is high, since the rel-
atively large trial displacements tend to equilibrate the local nonequilibrium
structures. At higher success rates, the trial displacements are small and the
swelling process tends to lock the particles into random structures. These
observations are reversed for the fraction of square defects (s
8
).
The s
6
and s
8
data for the thermodynamic and sheared structures are given in
Figs. 5.6 and 5.7, from which we observe the following:
In dense sheared systems, due to the sliding of hard-disk strings past each
other considerable concentration of square defects are generated in the sys-
tems. However at higher densities they get compacted out due to the lack of
geometric frustration, and the concentration of the crystallites tends towards
unity.
At a given packing fraction, as inelasticity increases, the concentration of the
square defects (s
8
) increases, with a consequent decrease in the concentration
Chapter 5 117
0.4 0.5 0.6 0.7 0.8 0.9
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0.9
1
PSfrag replacements
s
6
Figure 5.6: s
6
for hard-disks: thermodynamic structures (), sheared inelastic structures at
= 0.80 (), 0.90 (), 0.95 (+) and 0.99 (). Averaging as in Fig. 5.1.
0.4 0.5 0.6 0.7 0.8 0.9
0
0.05
0.1
0.15
0.2
0.25
0.3
PSfrag replacements
s
8
Figure 5.7: s
8
for hard-disks: thermodynamic structures (), sheared inelastic structures at
= 0.80 (), 0.90 (), 0.95 (+) and 0.99 (). Averaging as in Fig. 5.1.
of the crystallites (s
6
). This is the structural cause for the increase in shear
ordering packing fraction with increasing inelasticity.
The system size dependence check for sheared hard-disk structure f
6
, s
6
and s
8
are given in Table 5.1. Note that there is negligible size dependence for packing
fractions below the shear ordering packing fraction, and non-systematic size depen-
dence for the packing fractions above it, due to the intermittent ow of close-packed
hard-disk strings.
5.3 Hard-sphere structures
To every bond, formed between a sphere and its Voronoi neighbor, one can assign
a spherical harmonic Y
lm
(, ), where 0 and 0 < 2 are the angles
made by the bond with respect to some reference frame. The actual value of Y
lm
depends on the orientation of the reference frame, hence the second and third or-
der rotationally invariant combinations are used in the bond-orientational analysis,
Steinhardt et al. (1983).
Q
l
=
_
4
2 l + 1
l
m=l
[Y
lm
)[
2
_
1/2
, (5.3)
where Y
lm
) is the average over all the Voronoi bonds. Q
6
is lowest order Q
l
which
gives a nonzero value for the common crystalline structures, hence it is used as
an order parameter to sense any kind of crystallization, Steinhardt et al. (1983).
Richard et al. (1999) have used Q
6
to study the evolution of crystallization in the
random hard-sphere packings. Fig. 5.8 shows the Q
6
data for thermodynamic, ran-
dom and sheared structures, from which we note the following:
The freezing and melting packing fractions are respectively
F
0.494 and
M
0.545, Hoover and Ree (1968). The thermodynamic structures show a
sharp rise in Q
6
across the freezing transition. The dense random structures,
due to the absence of crystallization, do not show a rise in Q
6
.
The suppression of crystal nucleation in a homogeneous shear eld, Butler
and Harrowell (1995); Blaak et al. (2004), is evident in the Q
6
data. While the
Chapter 5 119
Table 5.1: System size dependence check for sheared inelastic hard-disk structure f
6
, s
6
and
s
8
, for a constant coef cient of restitution = 0.99.
Run f
6
s
6
s
8
0.60 I
a
0.6015 0.4347 0.2333
II
b
0.6022 0.4359 0.2326
III
c
0.6019 0.4356 0.2331
IV
d
0.6034 0.4368 0.2324
0.70 I 0.7628 0.6497 0.1274
II 0.7616 0.6488 0.1280
III 0.7624 0.6499 0.1280
IV 0.7624 0.6501 0.1278
0.75 I 0.8873 0.8375 0.0658
II 0.8881 0.8385 0.0653
III 0.8823 0.8284 0.0661
IV 0.8869 0.8355 0.0643
0.80 I 0.9830 0.9808 0.0143
II 0.9771 0.9716 0.0176
III 0.9533 0.9346 0.0296
IV 0.9518 0.9306 0.0273
a
Averaged for 10000 con gurations of 256 hard-disks, with PBC.
b
c
d
0 0.2 0.4 0.6
0
0.05
0.1
0.15
0.2
0.25
0.3
0.35
0.4
0.48 0.5 0.52 0.54 0.56 0.58
0
0.05
0.1
0.15
0.2
0.25
0.3
0.35
0.4
PSfrag replacements
Q
6
Figure 5.8: Q
6
for hard-spheres: thermodynamic structures (), random structures () gen-
erated at 50% success rate, and sheared inelastic structures at = 0.60 , 0.70 (), 0.80
(), 0.90 (+) and 0.99 (). Thermodynamic and random structures data averaged for 1000
con gurations of 256 hard-spheres, with PBC. Sheared structures data averaged for 675
con gurations of 384 hard-spheres, with PBC.
Chapter 5 121
0 0.5 1 1.5 2 2.5 3
x 10
3
0
2000
4000
6000
PSfrag replacements
f
(
Q
4
4
6
)
Figure 5.9: Distribution of Q
446
for hard-sphere thermodynamic structures at packing frac-
tions = 0.49 (thick line), 0.50 (), 0.55 (.), 0.60 and 0.65 (thin line).
thermodynamic ordering (i.e. the freezing) transition occurs at
F
0.494,
the shear ordering transition occurs in the range 0.52-0.53 for = 0.800.99,
in the range 0.54-0.55 for = 0.70 (details in the inset). Thus, the shear-
ordering packing fraction increases with increasing inelasticity.
For a local analysis, the local value of [Q
6
[ is not useful, since it has a nonzero
value for all the common crystalline structures. Hence, the third order rotational
invariant combinations are generally used, Steinhardt et al. (1983).
Q
l
1
l
2
l
3
=
m
1
+m
2
+m
3
=0
_
l
1
l
2
l
3
m
1
m
2
m
3
_
Q
l
1
m
1
Q
l
2
m
2
Q
l
3
m
3
, (5.4)
where () are the Wigner 3j symbols, the indices m
i
= l
i
, . . . , l
i
for i = 1, 2, 3,
and the summation is over the permutations satisfying the condition m
1
+m
2
+m
3
=
0. Mitus et al. (1995) have shown that Q
446
, dened on the 12 nearest neighbors
(note the metric, non-Voronoi denition of neighbors), can be used to identify the
local fcc structures. Gruhn and Monson (2001) have used this method to monitor
the fraction of fcc clusters formed during a hard-sphere solidication. With l
1
= 4,
0 0.5 1 1.5 2 2.5 3
x 10
3
0
500
1000
1500
2000
2500
3000
3500
4000
PSfrag replacements
f
(
Q
4
4
6
)
Figure 5.10: Distribution of Q
446
for sheared inelastic hard-spheres at a coef cient of resti-
tution = 0.90, at packing fractions = 0.52 (), 0.55 ( ), 0.60 (.) and 0.64
( ).
l
2
= 4 and l
3
= 6, there are 75 terms in the summation of Eq. 5.4. An ideal fcc
cluster has Q
446
2.5 10
3
. Fig. 5.9 shows the Q
446
distribution for the thermo-
dynamic structures. The Q
446
distribution has a two-peaked structure, which clearly
distinguishes the fcc structures from the non-fcc structures. Mitus et al. (1995) use
the cut-off Q
c
446
= 0.7 10
3
to distinguish the fcc clusters from other local struc-
tures. This classication is not sensitive to the actual cut-off value, since the two
peaks are well separated by a minimum. Fig. 5.10 shows the Q
446
distribution for
the sheared system. We observe that shear does not affect the distinctness of the
two-peaks. Hence, the criteria Q
446
> Q
c
446
can be used to identify fcc clusters,
even in the sheared congurations. The fraction of fcc clusters, f
fcc
, so counted
for the thermodynamic, random and sheared structures are given in Fig. 5.11, from
The observations regarding the thermodynamic freezing transition, the lack
of crystallization in random structures, and the suppression of crystal nucle-
Chapter 5 123
0.45 0.5 0.55 0.6 0.65
0
0.2
0.4
0.6
0.8
1
PSfrag replacements
f
f
c
c
Figure 5.11: f
fcc
for hard-spheres: thermodynamic structures (), random structures ()
generated at 50% success rate, and sheared inelastic structures at = 0.60 , 0.70 (),
0.80 (), 0.90 (+) and 0.99 (). Averaging as in Fig. 5.8.
Figure 5.12: An ideal bct cluster of 15 spheres (left) and the Voronoi polyhedron of the
central sphere in the cluster (right). The gures share the same view, but have different
scales.
ation in the sheared structures are identical with those noted in the Q
6
global
analysis above.
In the dense sheared structures, even in the dense and near-elastic limits, the
fraction of fcc clusters is only about 40%. However, the global Q
6
analysis in
Fig. 5.8 showed that the sheared structures are considerably ordered. This is
a clear indication of the coexistence of multiple crystalline structures (more
below).
We show the size dependence check for f
fcc
for the sheared structures in Table 5.2.
For packing fractions below the shear ordering packing fraction, there is negligible
size dependence. For packing fractions between shear ordering and the maximum
shearable packing fraction
m
0.641, Lun and Savage (1986), there is a mild but
non-systematic variation due to the intermittent sliding of close-packed layers past
each other.
Chapter 5 125
Table 5.2: System size dependence check for sheared inelastic hard-sphere structure f
fcc
, at
= 0.99.
Run f
fcc
100
0.45 I
a
3.26
II
b
3.26
III
c
3.33
IV
d
3.31
0.50 I 2.87
II 2.82
III 2.93
IV 2.96
0.55 I 40.47
II 35.22
III 46.52
IV 41.61
a
Averaged for 1200 con gurations of 216 (6 stacks of 6 6) hard-spheres, with PBC.
b
c
d
Before we proceed further with the microstructural analysis, we describe the
body-centered tetragonal structure. In the bct lattice, the lattice parameters are
a, a, c, with c ,= a in general. With c = a, the bct lattice reduces to the body-
centered cubic (bcc) lattice. A bct cluster contains 15 spheres, with six spheres
around the central sphere, forming a hexagonally close-packed plane, and four
spheres on either sides, Fig. 5.12. The four spheres above and below the central
layer are also close-packed, but these spheres instead of occupying the tetrahedral
voids offered by the central layer spheres (as in the close-packed fcc or hcp struc-
tures), are vertically above the midpoints of the central layer spheres. Hence, even
though the layers are close-packed, the stacking is not. In an ideal bct cluster, a
central sphere has 10 neighbors at a distance c = , and four spheres at a distance
a =
_
3/2, where is the hard-sphere diameter. The Voronoi polyhedron of
the central sphere in such a cluster is shown in Fig. 5.12. It has six square faces
(along the three orthogonal axis) and eight hexagonal faces. Visualizing only the
hexagon-forming spheres around the central sphere, reveals the cuboidal lattice. In
the cuboidal view, it is easily seen that the bct packing fraction is 2/9 0.6981.
For further details on tetrahedral packing see OKeeffe (1998) and the references
therein.
On shear ordering, close-packed planes of spheres slide past each other in a
zig-zag path, with the close-packed planes oriented parallel to the velocity-vorticity
plane. This phenomena is observed in the plastic deformation of fcc crystals (Cot-
trell (1953)), colloidal suspensions (Ackerson (1990)), and in the results of Chapter
4. When the close-packed layers slide thus past each other, the spheres in a layer
tend to occupy a position vertically above the centers of the spheres in the adja-
cent layers. This leads to the formation of bct structures. Fig. 5.13 shows a real
bct cluster realized in our simulation, compare it with the ideal cluster in Fig. 5.12.
This fcc-to-bct transition is the ow analogoue of the martensitic transformations
observed in rapidly quenched metals. In addition to the bct structure, we can expect
the in situ formation of hcp structure, due to the following reason: The fcc and the
hexagonal close packing (hcp) differ only in the stacking order, ABC and ABA
respectively. When close-packed planes slide past each other, the random stacking
faults form hcp structures. This phenomena is observed in the shear of colloidal
suspensions (Loose and Ackerson (1994)) and polymer solutions (Bang and Lodge
Chapter 5 127
Figure 5.13: A real bct cluster of 15 spheres formed during the homogeneous shear of
inelastic hard-spheres having a coef cient of restitution = 0.90, at a packing fraction
= 0.59.
(2003)). Gulley and Tao (1997) dened an order parameter to identify the forma-
tion of bct structure by electro-rheological suspensions in the presence of an electric
eld. This order parameter is not useful for our system due to the coexistence of
multiple crystalline structures. Hence we employ the pair analysis of Honeycutt
and Andersen (1987), which is widely used to study the microstructures in dense
amorphous systems, like metallic glasses (Duan et al. (2005)), amorphous alloys
(Pei et al. (2005)), coating suspensions (Qi et al. (1999)) etc.
In pair analysis, two particles are said form a bond if the distance between then
is less than a cut-off distance r
c
. Particles forming a bond are said to be neighbors.
Earlier applications of the pair analysis used a constant r
c
(1.38 for fcc and hcp
lattices and 1.54 for bcc lattice, in reduced units). However Yu et al. (2005) suggest
the rst minimum in the radial distribution function for fcc and hcp lattices and the
second minimum for bcc lattices. In this thesis we take the rst minimum in the
radial distribution function as r
c
. For sheared structures, we consider the angular
averaged radial distribution function. Some of the bonds occurring in dense particle
systems are illustrated in Fig. 5.14. Each pair of particles is characterized by four
PSfrag replacements
1441 1421 1422
1311 1551 1541
Figure 5.14: Some of the bonds observed in dense particle systems.
indices abcd. If two particles form a bond a = 1, otherwise a = 2. The number
of neighbors common to a given pair of particles is b. The number of bonds among
the common neighbors is c. The last index d is used to distinguish structures which
share same value for the rst three indices, especially to distinguish the 1421 and
1422 bonds. The general guidelines used to interpret the bonds are the following
(Lee et al. (2003)):
fcc structure has only 1421 bonds.
hcp structure has equal amounts of 1421 and 1422 bonds.
bcc structure has 43% 1441 and 57% 1661 bonds.
simple icosahedral structure has 71% 1321 and 29% 1551 bonds.
large icosahedral structures have signicant amounts of 1311 and 1422 bonds,
with a decrease in 1321 and 1551 bonds.
1431 and 1541 bonds are considered defective fcc and icosahedral structures,
Pei et al. (2005).
Chapter 5 129
0.45 0.5 0.55 0.6 0.65 0.7
0
0.2
0.4
0.6
0.8
1
PSfrag replacements
f
a
b
c
d
Figure 5.15: Fraction of bonds of the type abcd in thermodynamic hard-sphere structures.
abcd = 1421 (), 1422 (), 1441 (), 1661 (+), 1321 (), 1551 (), 1431 (), 1541 (),
and 1311 (). Averaging as in Fig. 5.8.
0.45 0.5 0.55 0.6 0.65
0
0.05
0.1
0.15
0.2
0.25
PSfrag replacements
f
a
b
c
d
Figure 5.16: Fraction of bonds of the type abcd in swelled random hard-sphere structures,
generated at 50% success rate. abcd = 1421 (), 1422 (), 1441 (), 1661 (+), 1321 (),
1551 (), 1431 (), 1541 (), and 1311 (). Averaging as in Fig. 5.8.
The pair-analysis results for the thermodynamic structures are given in Fig. 5.15.
A sharp rise in the 1421 bonds marks the onset of fcc order is hard-sphere solid
phase. For >
F
, the 1421 bonds increase at the expense of the defective fcc
and icosahedral structures, i.e. 1431 and 1541 bonds. The pair-analysis results for
the random structures are given in Fig. 5.16. The incidence of 1421 bonds, and
hence the formation of fcc crystallites is negligible. The dominant structures are
the icosahedra (1551) and the defective structures (1541 and 1431). We are not
aware of any work using pair-analysis to analyze the presence of bct structures. The
closest to the current thesis is that by Qi et al. (1999), which shows the coexistence
of fcc, hcp and bcc structures, along with icosahedral disorder, in sheared coating
suspensions. Hence, we have evolved the following additional rules by the visually
inspecting the local structures:
good bct clusters (as in Fig. 5.13) have nearly equal amounts of 1421 and
1422.
distorted bct clusters have higher amounts of 1421, 1431 and 1541 bonds.
These two observations can be supported as follows: Jakse et al. (2004) consider
the bonds 1441, 1431, 1421 and 1422 to indicate local tetrahedral order. This sup-
ports the rst observation. The dense amorphous structures, represented by icosahe-
dral structures containing 1551 and 1541 bonds, have local tetrahedral order, since
it maximizes the local packing. This supports the second observation. As men-
tioned earlier, when close-packed planes slide past each other, apart from bct struc-
tures, the hcp structures are also formed. Hence, we consider a preponderance of
1421 and 1422 bonds as a signature of both bct and hcp structures, and not exclu-
sively bct structure. The pair-analysis results for the sheared structures are given
in Fig. 5.17. Using the above additional rules, we interpret the results as follows:
in the high-dense limit, the sheared structures have about 60% 1421 and 20% 1422
bonds. Assuming that the bct+hcp structures are represented by equal amounts of
1421 and 1422 bonds, this implies that the fcc structure incidence is about 40%.
This estimate tallies with our earlier local analysis in Fig. 5.11. There is about 10%
of 1541 bonds and 8% of 1431, which could be defective bct structures. Note that
the incidence of bcc bonds (1441 and 1661) is only about 6%. In a bct structure
Chapter 5 131
0.45 0.5 0.55 0.6 0.65
0
0.1
0.2
0.3
0.4
0.5
0.6
0.7
PSfrag replacements
f
a
b
c
d
Figure 5.17: Fraction of bonds of the type abcd in sheared inelastic hard-sphere structures,
at a coef cient of restitution = 0.90 . abcd = 1421 (), 1422 (), 1441 (), 1661 (+),
1321 (), 1551 (), 1431 (), 1541 (), and 1311 (). Averaging as in Fig. 5.8.
c ,= a, the bcc-like bonds get counted as 1421 bonds and not as 1441 bonds. The
size dependence check for the sheared structure pair-analysis is given in Table 5.3.
There is a mild size dependence (< 3%): with increasing system size the fcc bonds
(1421) decreases, with an increase in icosahedral defects (1321 and 1311) and fcc
defects (1431). However, this trend does not counter our conclusion that bct+hcp
structures are coexisting with fcc structures. For packing fractions below the shear
ordering packing fraction, the system size dependence is negligible.
Nowwe present an analysis based the bcc-like polyhedra. Among the 14-faceted
polyhedra, consider the particular case of the polyhedra having six quadrilateral
and eight hexagonal faces. We call these polyhedra as the signature polyhedra,
since they are useful in identifying the local lattice parameters (shown below). The
neighbors of the signature polyhedron which form the six quadrilateral faces are
nearly mutually orthogonal (shown in the appendix). Hence, the local or pseudo
lattice parameters can be easily inferred from the distance between these neighbors
Table 5.3: System size dependence for the fraction of bond types in sheared inelastic hard-disk
structures at = 0.57 and = 0.90.
abcd I
a
II
b
III
c
IV
d
1421 0.3194 0.3036 0.3098 0.2862
1422 0.1392 0.1463 0.1287 0.1244
1441 0.0695 0.0693 0.0690 0.0696
1661 0.0598 0.0601 0.0598 0.0610
1321 0.0028 0.0035 0.0052 0.0093
1551 0.0184 0.0197 0.0189 0.0198
1431 0.1347 0.1374 0.1407 0.1476
1541 0.2347 0.2360 0.2352 0.2343
1311 0.0142 0.0157 0.0213 0.0309
others 0.0074 0.0084 0.0114 0.0168
a
b
c
d
1 1.1 1.2 1.3 1.4 1.5
1
1.1
1.2
1.3
1.4
1.5
PSfrag replacements
a/c
a
/
b
1 1.1 1.2 1.3 1.4 1.5
1
1.1
1.2
1.3
1.4
1.5
PSfrag replacements
a/c
a/b
a/c
a
/
b
Figure 5.18: The joint distribution of the ratios a/b and a/c at = 0.55 (left) and = 0.65
(right), for fcc hard-sphere solid structures. Darker region indicates higher probability.
Chapter 5 133
1 1.1 1.2 1.3 1.4 1.5
1
1.1
1.2
1.3
1.4
1.5
PSfrag replacements
a/c
a
/
b
1 1.1 1.2 1.3 1.4 1.5
1
1.1
1.2
1.3
1.4
1.5
PSfrag replacements
a/c
a/b
a/c
a
/
b
(right), for sheared inelastic hard-sphere structures at = 0.90.
1 1.1 1.2 1.3 1.4 1.5
1
1.1
1.2
1.3
1.4
1.5
PSfrag replacements
a/c
a
/
b
1 1.1 1.2 1.3 1.4 1.5
1
1.1
1.2
1.3
1.4
1.5
PSfrag replacements
a/c
a/b
a/c
a
/
b
(right), for hcp hard-sphere solid structures.
and the central sphere. Note that these pseudo lattice parameter are the actual lattice
parameters only for bcc and bct structure (shown below). In a bcc solid phase most
of the polyhedra are of this type. But in other solid structures, their incidence is
about 10 to 20%. We compute the local lattice parameters of the signature poly-
hedra as a, b, c, with a b c. We form the lattice ratios a/b and a/c, and
study their joint distribution. In the bcc lattice, both these ratios are unity. In the fcc
lattice, due to a topological instability, some of the second neighbors get promoted
as rst neighbors and form small quadrilateral face, Chapter 3. Such clusters form
the 14-faceted polyhedra with a/b = a/c =
2. Note that the actual fcc lattice

parameters are a = b = c. From Fig. 5.18, we observe that the thermodynamic
fcc solid phase evolves along the diagonal a/b = a/c and approaches the value
2
in the regular close-packed limit. In the sheared structures, the signature polyhe-
dra reveals the bct structure with the ratios a/b = 1 and a/c =
_
3/2 1.22.
In Fig. 5.19, at = 0.59 we observe that the joint distribution is centered around
the bct values for a shear ordered system. However, near the maximum shearable
packing fraction
m
0.641 (Lun and Savage (1986)), the peak position of the
joint distribution gets smeared towards the fcc value a/b = a/c =
2, since the
non-shearing plug retains the fcc structure. Thus, this polyhedra analysis shows the
bct signature in dense sheared systems. But it shares a shortcoming with the pair
analysis, that the bct and hcp structures respond identically. From Fig. 5.20, we
note that the hcp solid also has a/b = 1 and a/c =
_
3/2. The bcc-like signature
of the bct structure is found in the distribution of the number of faces (P
n
) of the
Voronoi polyhedra, as shown in Table 4.4. The fcc and hcp structures have polyhe-
dra with the number of faces in the range 12 to 18, due to the topological instability,
and share a nearly similarly distribution of faces. In bcc structures the 14-faceted
polyhedra are predominant. The predominance of the 14-faceted polyhedra in the
dense sheared inelastic structures shows the bcc-like signature of the bct structure.
Additionally a visual inspection, as in Fig. 5.13, conrms the bct signature in dense
sheared inelastic hard-sphere structures.
Chapter 5 135
5.4 Conclusions
We report the bond-orientational analysis results for the thermodynamic, random
and homogeneously sheared inelastic structures of hard-disks and hard-spheres.
The six-fold and four-fold orders in the hard-disk structures are monitored at dif-
ferent packing fractions. The thermodynamic structures contain negligible amount
of four-fold order and the six-fold order increases sharply across the freezing tran-
sition. The random hard-disk structures have low global six-fold order, yet consid-
erable local crystallization is observed due to the lack of geometric frustration in
two-dimensional systems. The local crystallization in the hard-disk random struc-
tures decreases with a decrease in the amplitude of the trial displacements involved
in their generation. Due to the suppression of crystal nucleation, the sheared struc-
tures get ordered at a packing fraction higher than the thermodynamic freezing
packing fraction. On shear ordering, strings of close-packed hard-disks, oriented
along the velocity direction, slide past each other. When close-packed strings of
disks slide past each other, considerable amount of four-fold order is created in the
system. This triangle-to-square transition is the two-dimensional ow analogue of
the martensitic transformations occurring in metals due to the stresses induced by a
rapid quench. While the transition in the metallic systems are due to diffusionless
lattice distortions occurring at rates comparable to the speed of sound, the transition
in the ow systems are due to the sliding of close-packed layers past each other
occurring over longer time scales.
In the hard-sphere structures, the global bond-orientational analysis shows the
onset of order in thermodynamic structures across the freezing transition, the ab-
sence of order in random structures and the suppression of nucleation in the homo-
geneously sheared structures. In sheared structures, even though the global analysis
shows a considerable level of order, a local analysis using a third-order rotational
invariant, devised by Mitus et al. (1995), shows that the fraction of spheres having
face-centered cubic (fcc) order is only about 40%, even in the dense and near-elastic
limits. Thus, the local analysis shows the presence of atleast one another crystalline
order in the sytem. On shear ordering, close-packed planes of spheres slide past
each other in a zig-zag path, with the close-packed planes oriented parallel to the
velocity-vorticity plane. When the close-packed layers slide thus past each other,
the spheres in a layer occupy a position vertically above the centers of the spheres in
the adjacent layers, forming the body-centered tetragonal (bct) structures. This fcc-
to-bct transition is the ow-analogoue of the martensitic transformations exhibited
by rapidly quenched metals. Also, when close-packed planes slide past each other,
the random stacking faults form hexagonal close-packed (hcp) structures. Using the
pair analysis of Honeycutt and Andersen (1987) and an analysis based on the 14-
faceted polyhedra having six quadrilateral and eight hexagonal faces, we show the
presence of bct and hcp signatures in shear ordered inelastic hard-spheres. Thus,
our analysis shows that the dense sheared inelastic hard-spheres have a mixture of
fcc, bct and hcp structures.
Appendix
5.A Orthogonality of the quadrilateral faces of the signature
polyhedra
A fraction of the 14-faceted polyhedra, have six quadrilateral and eight hexagonal
faces. We call these bcc-like polyhedra as the signature polyhedra, since it is easy to
deduce the local lattice parameters, as shown in Section 5.3. The near-orthogonality
of the neighbors contributing the quadrilateral faces is crucial to the quick estima-
tion of the local lattice parameters, and we demonstrate it here. Let r
i
, i = 1, . . . , 6,
be the unit vectors along the lines joining the central sphere and the neighbors con-
tributing the quadrilateral faces. We dene the non-orthogonality factor as
1 =
6
i=1
6
j=1
r
i
r
j
. (5.5)
This quantity is zero if the r
i
are mutually orthogonal. Because , for any i, the
contribution r
i
r
i
= 1 gets canceled by the contribution r
i
r
j
= 1, where
r
j
is the anti-parallel of r
i
, and the four remaining unit vectors lying in the plane
perpendicular to r
i
and r
j
make zero contributions. The outer summation in the
denition of 1 is necessary to locate the non-orthogonality among the unit vectors
lying in the plane normal to a given pair of unit vector and its anti-parallel. For slight
distortions of an orthogonal set r
i
, 1 has a small non-zero value. Hence, 1 is a
Chapter 5 137
Table 5.4: Percentage incidence and the orthogonality of the signature polyhedra in thermody-
namic and sheared inelastic hard-sphere structures.
Percentage R
Type incidence Minimum Average Maximum
thermodynamic
a
0.50 8.45 6.910
4
0.19 1.12
0.60 8.77 2.110
4
0.05 0.27
0.65 9.05 8.010
5
0.02 0.12
0.70 9.23 2.810
5
0.003 0.02
sheared
b
0.55 7.61 1.910
4
0.14 1.32
0.59 13.95 5.310
5
0.08 1.06
0.61 16.99 4.810
5
0.07 0.94
0.64 20.59 1.210
5
0.05 0.63
a
Averaged for 1000 con gurations of 256 hard-spheres, with PBC.
b
Averaged for 675 con gurations of 384 (6 stacks of 8 8) hard-spheres, with PBC, = 0.90.
measure of the non-orthogonality of the set r
i
. FromTable 5.4 we observe that the
range over which 1is distributed in the shear ordered structures is comparable with
that in the thermodynamic structures at = 0.50, which is just above the freezing
transition
F
0.494. Hence, the signature polyhedra of the sheared structures are
no more non-orthogonal than the just-frozen thermodynamic structures, in which
the fcc order has set in. Hence, we consider that the six neighbors forming the
quadrilateral faces in any signature polyhedra to be nearly mutually orthogonal.
Now, we separate the six neighbors contributing the quadrilateral faces into three
pairs of which lie along the same direction, i.e. the neighbor having a unit vector
r
i
and the neighbor having r
j
r
i
1. Half the distance between such a pair of
spheres give the lattice parameter along that direction.
References
Ackerson, B. J. (1990) Shear induced order and shear processing of model hard
sphere suspensions. J. Rheol. 34, 553.
Ackerson, B. J., Hayter, J. B., Clark, N. A. and Cotter, L. (1986) Neutron scattering
from charge stabilized suspensions undergoing shear. J. Chem. Phys. 84, 2344.
Alam, M. and Hrenya, C. M. (2001) Inelastic collapse in simple shear of a granular
medium. Phys. Rev. E 63, 061308.
Alder, B. J. and Wainwright, T. E. (1962) Phase transition in elastic disks. Phys.
Rev. 127, 359.
Alder, B. J. and Wainwright, T. E. (1967) Velocity autocorrelations for hard spheres.
Phys. Rev. Lett. 18, 988.
Allen, M. P. and Tildesley, D. J. (1992) Computer Simulation of Liquids. Clarendon,
Oxford.
Andrade, P. N. and Fortes, M. A. (1988) Distribution of cell volumes in a Voronoi
partition. Phil. Mag. B 58, 671.
Aurenhammer, F. (1991) Voronoi diagrams - a survey of a fundamental geometric
data structure. ACM Computing Surveys 23, 345.
Bang, J. and Lodge, T. P. (2003) Mechanisms and epitaxial relationships between
close-packed and bcc lattices in block copolymer solutions. J. Phys. Chem. B
107, 12071.
Barker, J. A. (1963) Lattice theories of the liquid state. Pergamaon Press, Oxford.
139
140 References
Barker, J. A. and Henderson, D. (1976) What is liquid? Understanding the states of
matter. Rev. Mod. Phys. 48, 587.
Barnes, H. A. (1989) Shear-thickening (Dilatancy) in suspensions of nonaggre-
gating solid particles dispersed in Newtonian liquids. J. Rheol. 33, 329.
Ben-Amotz, D. and Herschbach, D. R. (1990) Estimation of effective diameters for
molecular uids. J. Phys. Chem. 94, 1038.
Berryman, J. G. (1983) Random close packing of hard spheres and disks. Phys. Rev.
A 27, 1053.
Blaak, R., Auer, S., Frenkel, D. and Lowen, H. (2004) Crystal nucleation of col-
loidal suspensions under shear. Phys. Rev. Lett. 93, 068303.
Boltzmann, L. (1964) Lectures on Gas Theory, Translated by S. G. Brush, p. 169.
University of California Press, Berkley.
Bondi, A. (1954) Free volumes and free rotation in simple liquids and liquid satu-
rated hydrocarbons. J. Phys. Chem. 58, 929.
Boots, B. N. (1982) Arrangement of cells in random networks. Metallography 15,
53.
Bridges, F. G., Hatzes, A. and Lin, D. N. C. (1984) Structure, stability and evolution
of Saturns rings. Nature 309, 333.
Butler, S. and Harrowell, P. (1995) Kinetics of crystallization in a shearing colloidal
suspension. Phys. Rev. E 52, 6424.
Calka, P. (2003) An explicit expression for the distribution of the number of sides
of the typical Poisson-Voronoi cell. Adv. Appl. Prob. (SGSA) 35, 863.
Callister Jr., W. D. (2000) Materials Science and Engineering - An Introduction.
John Wiley & Sons, New York, fth edition.
Campbell, C. S. (1989) The stress tensor for simple shear ows of a granular mate-
rial. J. Fluid Mech. 203, 449.
References 141
Cao, J. and Ahmadi, G. (1995) Gas-particle two-phase turbulent ow in a vertical
duct. Int. J. Multiphase Flow 21, 1203.
Carnahan, N. F. and Starling, K. E. (1970) Thermodynamic properties of a rigid-
sphere uid. J. Chem. Phys. 53, 600.
Chandler, D., Weeks, J. D. and Andersen, H. C. (1983) Van der Waals picture of
liquids, solids and phase transformations. Science 220, 787.
Cohen, M. H. and Grest, G. S. (1979) Liquid-glass transition, a free-volume ap-
proach. Phys. Rev. B 20, 1077.
Cohen, M. H. and Grest, G. S. (1982) Erratum: Liquid-glass transition, a free-
volume approach. Phys. Rev. B 26, 6313.
Collins, R. (1968) A geometrical sum rule for two dimensional uid correlation
functions. J. Phys. C 1, 1461.
Consolati, G., Levi, M., Messa, L. and Tieghi, G. (2001) Free volumes and occupied
volumes in oligomeric polypropyleneglycols. Europhys. Lett. 53, 497.
Cottrell, A. H. (1953) Dislocations and Plastic Flow in Crystals. Clarendon, Ox-
ford.
Dahl, S. R. and Hrenya, C. M. (2004) Size segregation in rapid, granular ows with
continuous size distributions. Phys. Fluids 16, 1.
Deng, D., Argon, A. S. and Yip, S. (1989) A molecular dynamics model of melting
and glass transition in an idealized two-dimensional material I. Phil. Trans. R.
Soc. Lond. A 329, 549.
DiCenzo, S. B. and Wertheim, G. K. (1989) Monte Carlo calculation of the size
distribution of supported clusters. Phys. Rev. B 39, 6792.
Doolittle, D. K. (1951) Studies in Newtonian ow. II. The dependence of the vis-
cosity of liquids on free-space. J. App. Phys. 22, 1471.
Drouffe, J. M. and Itzykson, C. (1984) Random geometry and the statistics of two-
dimensional cells. Nuc. Phys. B 235, 45.
142 References
Duan, G., Xu, D., Zhang, Q., Zhang, G., Cagin, T., Johnson, W. L. and Goddard III,
W. A. (2005) Molecular dynamics study of the binary Cu
46
Zr
54
metallic glass
motivated by experiments: Glass formation and atomic level structure. Phys.
Rev. B 71, 224208.
Englman, R., Rivier, N. and Jaeger, Z. (1987) Fragment-size distribution in disinte-
gration by maximum-entropy formalism. Phil. Mag. B 56, 751.
Finney, J. L. (1970) Random packings and the structure of simple liquids I. The
geometry of random close packing. Proc. Roy. Soc. London A 319, 479.
Fisher, I. Z. (1964) Statistical Theory of Liquids. The University of Chicago Press,
Chicago.
Foerster, S. F., Louge, M. Y., Chang, H. and Allia, K. (1994) Measurements of the
collision properties of small spheres. Phys. Fluids 6, 1108.
Fuller, R. B. (1975) Synergetics - Explorations in the Geometry of Thinking.
Macmillan Publishing Co. Inc., New York.
Glaser, M. A. and Clark, N. A. (1993) Melting and liquid structure in two dimen-
sions. Adv. Chem. Phys. 83, 543.
Goldhirsch, I. and Tan, M. L. (1996) The single-particle distribution function for
rapid granular shear ows of smooth inelastic disks. Phys. Fluids 8, 1752.
Gotoh, K. (1993) Comparison of random congurations of equal disks. Phys. Rev.
E 47, 316.
Gradshteyn, I. S. and Ryzhik, I. M. (2000) Table of Integrals, Series and Products.
Academic Press, San Diego, USA, 6th edition.
Grest, G. S. and Cohen, M. H. (1981) Liquids, glasses, and the glass transition: a
free-volume approach. Adv. Chem. Phys. 48, 455.
Gruhn, T. and Monson, P. A. (2001) Molecular dynamics simulations of hards
sphere solidication at constant pressure. Phys. Rev. E 64, 061703.
References 143
Gulley, G. L. and Tao, R. (1997) Structures of an electrorheological uid. Phys.
Rev. E 56, 4328.
Halperin, B. I. and Nelson, D. R. (1978) Theory of two-dimensional melting. Phys.
Rev. Lett. 41, 121.
Hanson, H. G. (1983) Voronoi cell properties from simulated and real random
spheres and points. J. Stat. Phys. 30, 591.
Haward, R. N., ed. (1973) The Physics of Glassy Polymers. Applied Science Pub-
lishers, London.
Hayakawa, H. and Kuninaka, H. (2002) Simulation and theory of two-dimensional
elastic disks. Chem. Engg. Sc. 57, 239.
Hermann, H., Wendrock, H. and Stoyan, D. (1989) Cell-Area Distributions of Pla-
nar Voronoi Mosaics. Metallography 23, 189.
Heyes, D. M. (1988) Transport coefcients of Lennard-Jones uids: a molecular-
dynamics and effective-hard-sphere treatment. Phys. Rev. B 37, 5677.
Hill, T. L. (1956) Statistical Mechanics. McGraw-Hill Book Company, Inc., New
York.
Hinde, A. L. and Miles, R. E. (1980) Monte Carlo estimates of the distributions of
the randompolygons of the Voronoi tessellation with respect to a Poisson process.
J. Statist. Comput. Simul. 10, 205.
Hinrichsen, E. L., Feder, J. and Jossang, T. (1990) Random packing of disks in two
dimensions. Phys. Rev. A 41, 4199.
Hoffman, R. L. (1972) Discontinuous and dilatant viscosity behavior in concen-
trated suspensions. I. Observations of a ow instability. Trans. Soc. of Rheol. 16,
155.
Hoffman, R. L. (1998) Explanations for the cause of shear thickening in concen-
trated colloidal suspensions. J. Rheol. 42, 111.
144 References
Honeycutt, J. D. and Andersen, H. C. (1987) Molecular dynamics study of melting
and freezing of small Lennard-Jones clusters. J. Phys. Chem. 91, 4950.
Hoover, W. G. (1966) Pressure and entropy for hard particles at high density. J.
Chem. Phys. 44, 221.
Hoover, W. G., Hoover, N. E. and Hanson, K. (1979) Exact hard disk free-volumes.
J. Chem. Phys. 70, 1837.
Hoover, W. G. and Ree, F. H. (1968) Melting transition and communal entropy for
hard spheres. J. Chem. Phys. 49, 3609.
Hsu, C. S. and Rahman, A. (1979) Interaction potentials and their effect on crystal
nucleation and symmetry. J. Chem. Phys. 71, 4974.
Ikeda, H., Qi, Y., Samwer, K., Johnson, W. L. and Goddard III, W. A. (1999) Strain
rate induced amorphization in metallic nanowires. Phys. Rev. Lett. 82, 2900.
Jakse, N., Lebacq, O. and Pasturel, A. (2004) Ab initio molecular dynamics simu-
lations of short-range order in liquid Al
80
Mn
20
and Al
80
Ni
20
alloys. Phys. Rev.
Lett. 93, 207801.
Jaynes, E. T. (1957) Information theory and statistical mechanics. Phys. Rev. 106,
620.
Jaynes, E. T. (1965) Gibbs vs Boltzmann entropies. Am. J. Phys. 33, 391.
Jaynes, E. T. (1979) Where do we stand on maximum entropy? In The Maximum
Entropy Formalism, eds. Levine, R. D. and Tribus, M. The MIT Press, Cam-
bridge, MA. Available at http://bayes.wustl.edu/.
Johnson, K. L. (1985) Contact Mechanics. Cambridge University Press, New York.
Jullien, R., Jund, P., Caprion, D. and Quitmann, D. (1996) Computer investigation
of long-range correlations and local order in random packings of spheres. Phys.
Rev. E 54, 6035.
Kanai, T., Sawada, T., Maki, I. and Kitamura, K. (2003) Kossel line analysis of
ow-aligned textures of colloidal crystals. Jpn. J. Appl. Phys. 42, L655.
References 145
Kiang, T. (1966) Random fragmentation in two and three dimensions. Z. Astrophys.
64, 433.
Kirkwood, J. G. (1950) Critique of the free volume theory of the liquid state. J.
Chem. Phys. 18, 380.
Kirkwood, J. G. and Boggs, E. M. (1942) The radial distribution function in liquids.
J. Chem. Phys. 10, 394.
Kumar, V. S. and Kumaran, V. (2005a) Voronoi cell-volume distribution and con-
gurational entropy of hard-spheres. J. Chem. Phys. 123, 114501.
Kumar, V. S. and Kumaran, V. (2005b) Voronoi neighbor statistics of hard-disks and
hard-spheres. J. Chem. Phys. 123, 074502.
Kumar, V. S. and Kumaran, V. (2006a) Bond-orientational analysis of hard-disk and
hard-sphere structures. Phys. Rev. E under review.
Kumar, V. S. and Kumaran, V. (2006b) Voronoi neighbor statistics of homoge-
neously sheared inelastic hard-disks and hard-spheres. Phys. Rev. E accepted
for publication.
Kuwabara, G. and Kono, K. (1987) Restitution coefcient in a collision between
two spheres. Jpn. J. Appl. Phys., Part 1 26, 1230.
Labous, L., Rosato, A. D. and Dave, R. N. (1997) Measurements of collisional
properties of spheres using high-speed video analysis. Phys. Rev. E 56, 5717.
Lado, F. (1984) Choosing the reference system for liquid state perturbation theory.
Mol. Phys. 52, 871.
Lavrik, N. L. and Voloshin, V. P. (2001) Calculation of mean distances between
the randomly distributed particles in the model of points and hard spheres (the
method of Voronoi polyhedra). J. Chem. Phys. 114, 9489.
Lee, H. J., Cagin, T., Johnson, W. L. and Goddard III, W. A. (2003) Criteria for
formation of metallic glasses: the role of atomic size ratio. J. Chem. Phys. 119,
9858.
146 References
Lees, A. W. and Edwards, S. F. (1972) The computer study of transport processes
under extreme conditions. J. Phys. C 5, 1921.
Lepori, L. and Gianni, P. (2000) Partial molar volumes of ionic and nonionic or-
ganic solutes in water: A simple additivity scheme based on the intrinsic volume
approach. J. Sol. Chem. 29, 405.
Levelt, J. M. H. and Cohen, E. G. D. (1964) In Studies in Statistical Mechanics,
Vol.II, Part B, eds. Boer, J. D. and Uhlenbeck, G. E. North-Holland Publishing
Company, Amsterdam.
Li, L. Y., Wu, C. Y. and Thornton, C. (2002) A theoretical model for the contact of
elastoplastic bodies. J. Mechanical Engineering Science 216C, 421.
Loose, W. and Ackerson, B. J. (1994) Model calculations for the analysis of scat-
tering data from layered structures. J. Chem. Phys. 101, 7211.
Lun, C. K. K. and Savage, S. B. (1986) The effects of an impact velocity depen-
dent coefcient of restitution on stresses developed by sheared granular materials.
Acta Mechanica 63, 15.
Maeso, M. J. and Solana, J. R. (1993) Instabilities in the equations of state of hard
disk and hard sphere uids from the virial expansion. J. Chem. Phys. 99, 548.
Matsuoka, S. and Hale, A. (1997) Cooperative relaxation processes in polymers. J.
App. Poly. Sc. 64, 77.
Mazenko, G. F. (2000) Equilibrium Statistical Mechanics. Wiley-VCH, Berlin.
McGowan, J. C. and Mellors, A. (1986) Molecular Volumes in Chemistry and Biol-
ogy Applications Including Partitioning and Toxicity. Ellis Harwood, Chichester,
UK.
McNamara, S. and Falcon, E. (2005) Simulations of vibrated granular medium with
impact-velocity-dependent restitution coefcient. Phys. Rev. E 71, 031302.
Meijering, J. L. (1953) Interface area, edge length, and number of vertices in crystal
aggregates with random nucleation. Phillips Res. Rep. 8, 270.
References 147
Mitus, A. C., Smolej, F., Hahn, H. and Patashinski, A. Z. (1995) Q
446
shape spec-
troscopy of local f.c.c. structures in computer simulations of crystallization. Eu-
rophys. Lett. 32, 777.
Nishiyama, Z. (1978) Martensitic Transformations. Academic Press, New York.
Ogawa, T. and Tanemura, M. (1974) Geometrical considerations on hard core prob-
lems. Prog. Theo. Phys. 51, 399.
Oger, L., Gervois, A., Troadec, J. P. and Rivier, N. (1996) Voronoi tessellation of
packings of spheres: topological correlations and statistics. Phil. Mag. B 74, 177.
Okabe, A., Boots, B. and Sugihara, K. (1992) Spatial tessellations: Concepts and
Applications of Voronoi Diagrams. Wiley, New York.
OKeeffe, M. (1998) Sphere packings and space lling by congruent simple poly-
hedra. Acta. Cryst. A 54, 320.
Onoda, G. Y. and Liniger, E. G. (1990) Random loose packings of uniform spheres
and the dilatancy onset. Phys. Rev. Lett. 64, 2727.
ORourke, J. (1994) Computational Geometry in C. Cambridge University press,
Cambridge.
Pei, Q. X., Lu, C. and Lee, H. P. (2005) Crystallization of amorphous alloy during
isothermal annealing: a molecular dynamics study. J. Phys.: Condens. Matter
17, 1493.
Poschel, T., Brilliantov, N. V. and Schwager, T. (2003) Long-time behavior of gran-
ular gases with impact-velocity dependent coefcient of restitution. Physica A
325, 274.
Poschel, T. and Schwager, T. (1998) Comment on Clustering of granular assem-
blies with temperature dependent restitution under Keplerian differential rota-
tion. Phys. Rev. Lett. 80, 5708.
Preparata, R. S., Preparata, F. P. and Shamos, M. I. (1993) Computational Geome-
try. Springer-Verlag, New York.
148 References
Qi, D., Joyce, M. and Fleming, D. (1999) Analysis of microstructure of coating
suspensions. Powder Technology 104, 50.
Quickenden, T. I. and Tan, G. K. (1974) Random packing in two dimensions and
the structure of monolayers. J. Colloid and Interface Sci. 48, 382.
Rahman, A. (1966) Liquid structure and self-diffusion. J. Chem. Phys. 45, 2585.
Raman, C. V. (1918) The photographic study of impact at minimal velocities. Phys.
Rev. 12, 442.
Rapaport, D. C. (1983) Density uctuations and hydrogen bonding in supercooled
water. Mol. Phys. 48, 23.
Reiss, H. (1992) Statistical geometry in the study of uids and porous media. J.
Phys. Chem. 96, 4736.
Richard, P., Oger, L., Troadec, J. P. and Gervois, A. (1999) Geometrical characteri-
zation of hard-sphere systems. Phys. Rev. E 60, 4551.
Sanchez, I. C. (1994) Virial coefcients and close-packing of hard spheres and
disks. J. Chem. Phys. 101, 7003.
Savage, S. B. and Jeffrey, D. J. (1980) In Euromech, volume 133. Oxford Univ.
Schliecker, G. (2002) Structure and dynamics of cellular systems. Adv. in Phys. 51,
1319.
Schwager, T. and Poschel, T. (1998) Coefcient of normal restitution of viscous
particles and cooling rate of granular gases. Phys. Rev. E 57, 650.
Shannon, C. E. (1948) A mathematical theory of communica-
tion. Bell Sys. Tech. J. 27, 379. Available at http://cm.bell-
labs.com/cm/ms/what/shannonday/paper.html.
Sierou, A. and Brady, J. F. (2002) Rheology and microstructure of concentrated
noncolloidal suspensions. J. Rheol. 46, 1031.
References 149
Silva, C., Liu, H. and Macedo, E. A. (1998) Comparison between different explicit
expression of the effective hard sphere diameter of Lennard-Jones uid: applica-
tion to self-diffusion coefcients. Ind. Eng. Chem. Res. 37, 221.
Spahn, F., Schwarz, U. and Kurths, J. (1997) Clustering of granular assemblies with
temperature dependent restitution under Keplerian differential rotation. Phys.
Rev. Lett. 78, 1596.
Spanier, J. and Oldham, K. B. (1987) An Atlas of Functions. Hemisphere Publishing
Corp., Washington.
Speedy, R. J. (1987) Diffusion in the hard sphere uid. Mol. Phys. 62, 509.
Stancik, E. J., Gavranovic, G. T., Widenbrant, M. J. O., Laschitsch, A. T., Vermant,
J. and Fuller, G. G. (2003) Structure and dynamics of particle monolayers at a
liquid-liquid interface subjected to shear ow. Faraday Discuss. 123, 145.
Steinhardt, P. J., Nelson, D. R. and Ronchetti, M. (1983) Bond-orientational order
in liquids and glasses. Phys. Rev. B 28, 784.
Stevens, M. J. and Robbins, M. O. (1993) Simulations of shear-induced melting and
ordering. Phys. Rev. E 48, 3778.
Stillinger, D. K., Stillinger, F. H., Torquato, S., Truskett, T. M. and Debenedetti,
P. G. (2000) Triangle distribution and equation of state for classical rigid disks.
J. Stat. Phys. 100, 49.
Tanemura, M. (2001) Random Voronoi cells in higher dimensions. Visual Math. 3.
Available at http://www.mi.sanu.ac.yu/vismath/proceedings/tanemura.htm.
Tanemura, M. (2003) Statistical distributions of Poisson Voronoi cells in two and
three dimensions. Forma 18, 221.
Tanemura, M., Hiwatari, Y., Matsuda, H., Ogawa, T., Ogita, N. and Ueda, A. (1977)
Geometrical analysis of crystallization of the soft-core model. Prog. Theor. Phys.
58, 1079.
Tonks, L. (1936) The complete equation of state of one, two and three-dimensional
gases of hard elastic spheres. Phys. Rev. 50, 955.
150 References
Torquato, S., Truskett, T. M. and Debenedetti, P. G. (2000) Is random close packing
of spheres well dened? Phys. Rev. Lett 84, 2064.
Troadec, J. P., Gervois, A. and Oger, L. (1998) Statistics of Voronoi cells of slightly
perturbed face-centered cubic and hexagonal close-packed lattices. Europhys.
Lett. 42, 167.
van Rensburg, E. J. J. (1993) Virial coefcients for hard discs and hard spheres. J.
Phys. A 26, 4805.
Verlet, L. (1968) Computer experiments on classical uids II. Equilibrium correla-
tion functions. Phys. Rev. 165, 201.
Vermant, J. and Solomon, M. J. (2005) Flow-induced structure in colloidal suspen-
sions. J. Phys: Condens. Matter 17, R187.
Voronoi, G. (1908) Nouvelles applications des parametres continus a la theorie des
formes quadratiques. J. Reine Angew. Math. 134, 198.
Weaire, D., Kermode, J. P. and Wejchert, J. (1986) On the distribution of cell areas
in a Voronoi network. Phil. Mag. B 53, L101.
Weiss, J. A., Oxtoby, D. W. and Grier, D. G. (1995) Martensitic transition in a
conned colloidal suspension. J. Chem. Phys. 103, 1180.
Wood, W. W. (1952) Note on the free volume equation of state for hard spheres. J.
Chem. Phys. 20, 1334.
Wood, W. W. (1967) Monte Carlo calculations for hard disks in the isothermal-
isobaric ensemble. J. Chem. Phys. 48, 415.
Woodcock, L. V. (1976) Glass transition in the hard sphere model. J. Chem. Soc.
Farad. Trans. II 72, 1667.
Woodcock, L. V. (1985) Origins of thixotropy. Phys. Rev. Lett. 54, 1513.
Yamamoto, T. and Furukawa, H. (2001) Relationship between molecular structure
and zero-shear viscosity of polymers. J. App. Poly. Sci. 80, 1609.
References 151
Young, D. A. and Alder, B. J. (1979) Studies in molecular dynamics. XVII. Phase
diagrams for step potentials in two and three dimensions. J. Chem. Phys. 70,
473.
Yu, D. Q., Chen, M. and Han, X. J. (2005) Structure analysis methods for crystalline
solids and supercooled liquids. Phys. Rev. E 72, 051202.
Zhu, H. X., Thorpe, S. M. and Windle, A. H. (2001) The geometrical properties of
irregular two-dimensional Voronoi tessellations. Phil. Mag. A 81, 2765.

PHD Thesis

Uploaded by

Document Information

Original Title

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

PHD Thesis

Uploaded by

Copyright:

Available Formats

Statistical geometric analysis of

hard-disk and hard-sphere

0 and the plastic

is the volume swept by the center

m) of the cell volume

form the spanning cluster and (N

= /(2D), where D is the dimensionality of the system. Using

values from the models in Table 3.2 Voronoi analysis

BH WCA LWCA BAH C

0.7408 0.8350 1.0262 1.0224 1.0200 1.0123 14.47 1.0318

2) is the regular close-packed packing fraction.

3/2) in the hexagonal

2. Note that the actual fcc lattice

You might also like